diff --git a/README.md b/README.md
index 46622790..dfe5d0d7 100644
--- a/README.md
+++ b/README.md
@@ -15,25 +15,53 @@
---
-DeepAgent Code is an AI coding agent built on persistent document memory. It keeps [opencode](https://github.com/sst/opencode)'s runtime foundations and adds a control plane for **durable knowledge**, **cross-session memory**, **context assembly**, **learning lifecycle**, and **runtime intelligence**.
+DeepAgent Code is an AI coding agent built on persistent document memory. It keeps [opencode](https://github.com/sst/opencode)'s runtime foundations and adds a control plane so the agent behaves less like a one-shot chat and more like a teammate that remembers your project, sharpens vague asks, and goes deep on hard problems.
-## What Makes It Different
+The features below start from a real need — something a plain coding agent, opencode included, leaves on the table — and work down to the architecture we built to serve it.
-**Persistent document system** — Knowledge, decisions, diagnostics, and learnings are stored as typed documents in a searchable graph. The agent builds understanding across sessions instead of starting from scratch each time.
+## What You Can Do
-**Project memory sharing** — Multiple conversations within the same project share knowledge, coding patterns, common pitfalls, and build commands. What one session learns becomes available to the next.
+### Keep one conversation going indefinitely
-**AI IDE microservice** — Query code by symbol name and intent (not file:line coordinates). Get definitions, references, call chains, type hierarchies, and diagnostics in one call. Built on LSP with 38 language servers.
+**The need:** Long tasks overflow the context window. Most agents respond by truncating history or summarizing everything at a threshold — so mid-task the agent forgets a decision you made an hour ago, or the window fills with stale tool output and quality falls off a cliff.
-**Preset MCP catalog** — Curated MCP servers for Git platforms, file search, read-only databases, and browser automation. Risk tiers derived at runtime from catalog structure, not user config.
+**What DeepAgent does:** Your conversation is treated as a continuously maintained work state, not a growing chat log. Before every turn, the agent rebuilds a working set — the task anchor, the most recent exchanges verbatim, active file references, and only the older facts that are relevant right now — while the full history is archived durably and stays queryable. The working set is held to a hard fraction of the model window, so there's always room for the model to actually think and respond. You just keep talking; the agent keeps focus.
-**Domain packs** — Specialized knowledge packages for specific domains (GPU kernels, React, backend APIs, security, testing). Each pack contains typed documents (strategies, methodologies, knowledge, skills) + validation/diagnostic adapters. Packs compose: activate multiple domains for your task.
+### Switch windows, fork freely, never lose memory
-**Learning lifecycle** — After completing work, the agent can generate candidate memories, facts, strategies, and methodologies. Evidence and approval gates control what gets persisted.
+**The need:** You want to try two approaches, or hand the work to a fresh conversation, without starting from amnesia and without polluting the original thread.
-**Work strength ladder** — `general`, `high`, `xhigh`, `max`, `ultra` scale capability without breaking contracts. Higher strengths add control-plane abilities (multi-agent orchestration, adversarial validation) on top of base behavior.
+**What DeepAgent does:** Fork any conversation from a chosen message. The fork opens carrying the parent's memory up to that point, shows a full-width "derived from" marker at the top of its transcript, and nests folder-style under its origin in the session tree (subagents and forks alike, up to three levels deep). Knowledge flows up a scope hierarchy — what one session learns can be promoted to the whole project, and cross-project preferences live at the user-global layer — so switching windows is a clean handoff, not a reset.
-**Scenario modes** — `direct` executes immediately; `wish` refines intent first, shows draft plan, waits for confirmation before automation.
+### Ask roughly, and let the agent sharpen it
+
+**The need:** A half-formed prompt gets a half-useful answer, and you don't always know how to phrase what you want.
+
+**What DeepAgent does:** Two scenario modes on the composer. **Direct** sends your prompt as-is — you own the wording. **Intelligence** refines a rough ask into a sharper prompt, surfaces a draft plan and decision suggestions, and waits for your confirmation before it automates anything. You decide how much the agent shapes the request.
+
+### Go deep on genuinely hard problems
+
+**The need:** Complex work — an architecture decision, a tricky migration, a subtle bug — needs more than a single confident pass. It needs research, a second opinion, and someone actively trying to poke holes.
+
+**What DeepAgent does:** At higher work strengths the primary agent decomposes the task, fans it out to focused subagents that research modules in parallel, synthesizes their findings, and then runs independent reviewers whose job is to *break* the plan rather than agree with it. Fan-out is bounded by a configurable concurrency ceiling, and live subagents surface in a session side panel and inline in the transcript so you can watch and jump into any of them.
+
+### Chat with your team and your agents in one place
+
+**The need:** Coordinating with teammates and driving agents usually happens in two different tools.
+
+**What DeepAgent does:** A per-project group chat lives in the session side panel. @mention an agent as a chat member and it runs the full agent loop — query code, generate, fix — pulling project knowledge and recent messages for context, then replies inline with live progress streaming.
+
+## How It Works
+
+Each capability above is served by a control-plane primitive underneath. These are the parts a plain runtime doesn't have.
+
+**Four-graph unification** — Code, knowledge, project memory, and the document graph are unified into one typed, bidirectionally-linked store. When the agent pulls context, a change to a symbol surfaces the design decisions, past diagnoses, and knowledge actually linked to it — connected context, not four disconnected keyword searches.
+
+**Domain packs** — 140+ composable knowledge packages spanning languages, frameworks, platforms (cloud, Kubernetes, CI), hardware, and business/risk domains (security, privacy, compliance). Each pack bundles typed documents (strategies, methodologies, knowledge, skills, failure dossiers) with detectors that auto-activate the right packs for your task; conflicts resolve stricter-policy-wins, and the active set is version-locked so a run is reproducible. Core stays domain-neutral — expertise is data on disk, not hardcoded.
+
+**Tiered knowledge invocation** — A monotonic strength ladder (`general → high → xhigh → max → ultra`) gates how much control-plane machinery engages. `general` stays close to the plain runtime — fast and cheap. Higher rungs progressively unlock durable knowledge, project handoff summaries, heavier strategy/methodology tiers, and multi-agent orchestration. You pay for depth only when you dial it up.
+
+**Self-learning** — After work lands, the agent proposes candidate knowledge, facts, and methodologies. Promotion is evidence-gated (a test passed, a diagnostic cleared, a validation confirmed) and user-controllable — durable knowledge is carried over deliberately, not silently guessed. Session-stable conclusions consolidate into project memory over time, so the next session starts smarter about *your* codebase.
## Installation
@@ -61,43 +89,41 @@ The agent will:
1. Use LSP to find the endpoint definition and understand its structure
2. Check project memory for existing middleware patterns
-3. Implement rate limiting following project conventions
-4. Run tests and capture diagnostics
-5. Generate a candidate memory: "This project uses express-rate-limit middleware"
+3. Activate the relevant domain packs (backend API, the project's language)
+4. Implement rate limiting following project conventions
+5. Run tests, capture diagnostics, and propose a candidate memory: "This project uses express-rate-limit middleware"
On your next session, when you ask to add rate limiting elsewhere, the agent already knows the pattern.
## Core Concepts
-**Document graph** — All persistent state lives in typed documents: `knowledge`, `strategy`, `methodology`, `skill`, `memory`, `design`, `worklog`, `diagnosis`, `eval`. Documents link to each other (supports/blocks/conflicts/validates), forming a graph.
+**Document graph** — All persistent state lives in typed documents: `knowledge`, `strategy`, `methodology`, `skill`, `memory`, `design`, `worklog`, `diagnosis`, `eval`. Documents link to each other (supports/blocks/conflicts/validates), forming a graph you can traverse.
**Scope layers** — `session-private` (current conversation), `project-shared` (all sessions in this project), `user-global` (cross-project preferences), `public-system` (built-in skills), `sealed` (audit-only, never enters context).
-**Domain packs** — Each pack (e.g., `code.frontend.react`, `code.gpu-kernel`, `risk.security`) is a bundle of typed documents + adapters. Documents include strategies (directions), methodologies (multi-step workflows), knowledge (facts), skills (executable capabilities), and failure dossiers. Packs auto-activate based on problem profile or explicit selection. Core stays domain-neutral.
-
-**Context admission** — Retrieval hits go through admission gates. Sensitive information (SSH hosts, tokens, internal paths) gets suggested but not auto-expanded into prompts.
+**Context admission** — Retrieval hits pass through admission gates. Full tool output (raw LSP dumps, diagnostics, capability indexes) is written to evidence artifacts, ref-linked and tool-only; only summaries and `file:line` snippets enter the model context. Sensitive values (SSH hosts, tokens, internal paths) are suggested, never auto-expanded.
-**Evidence-gated learning** — Learnings require evidence (test pass, diagnostic clear, validation confirmed). Candidates enter a queue; auto-merge or manual review depends on policy.
+**AI IDE microservice** — Query code by symbol name and intent (e.g. `code_intel({ symbol: "AgentGateway.open", intent: "overview" })`), not file:line coordinates. Get definitions, references, call chains, type hierarchies, and diagnostics in one call. Built on LSP with 38 language servers; degrades gracefully to grep/read when no server is configured.
-**Symbol-driven navigation** — Code intelligence tools accept symbol names (e.g., "AgentGateway.open"), not coordinates. The agent resolves names to locations internally via LSP workspace/document symbols.
+**Preset MCP catalog** — Curated MCP servers for Git platforms, file search, read-only databases, and browser automation. Risk tiers are derived at load time from the catalog template (not user config, so they can't be injected), and servers default to not-connected with write and external-fetch operations behind approval gates.
## Architecture
```
┌─────────────────────────────────────────────────────────────┐
│ Control Plane (DeepAgent additions) │
-│ • Document graph (persistent memory) │
+│ • Four-graph unified store (code + knowledge + memory + doc)│
+│ • Continuously maintained working state (memory + compaction)│
+│ • Domain pack system (composable, auto-activating knowledge)│
│ • Context assembly & admission gates │
-│ • Learning worker (background, non-blocking) │
-│ • Evidence & approval gates │
-│ • Work strength orchestration │
-│ • Domain pack system (composable knowledge) │
+│ • Multi-agent orchestration & adversarial review │
+│ • Evidence-gated learning & work-strength ladder │
└─────────────────────────────────────────────────────────────┘
│
┌─────────────────────────────────────────────────────────────┐
│ Runtime Foundations (from opencode) │
│ • Agent loop & tool execution │
-│ • Session & provider management │
+│ • Session, fork & provider management │
│ • MCP client runtime │
│ • Permission system │
└─────────────────────────────────────────────────────────────┘
@@ -111,7 +137,7 @@ On your next session, when you ask to add rate limiting elsewhere, the agent alr
└─────────────────────────────────────────────────────────────┘
```
-DeepAgent's control plane operates at provider-turn boundaries: it selects context before each model call and writes evidence back into the document graph afterward. It does not replace opencode's runtime—it layers on top.
+DeepAgent's control plane operates at provider-turn boundaries: it selects context before each model call and writes evidence back into the document graph afterward. It does not replace opencode's runtime — it layers on top.
## Documentation
@@ -127,22 +153,6 @@ DeepAgent Code is licensed under **AGPL-3.0-or-later**. If you modify and run it
This project is derived from [opencode](https://github.com/sst/opencode) (MIT License). See [NOTICE](NOTICE) for the upstream license and attribution. No endorsement by opencode or its contributors is implied.
-## Project Status
-
-**V3.4.1** is the first public pre-release hardening milestone. It includes:
-
-- LSP-to-AI-IDE transformation (symbol-driven code intelligence)
-- Preset MCP catalog with security model
-- License and attribution cleanup
-- Secret scan baseline
-- Design documentation consolidation
-
-**V3.5** (planned) will add:
-
-- DAP integration (debug adapter protocol for runtime intelligence)
-- PAP (performance analysis protocol for profiling: NVIDIA NCU/nsys, AMD rocprof, Intel VTune, CPU perf)
-- OS-backed credential storage for MCP servers
-
---
+ )
+}
diff --git a/packages/app/src/components/im/group-chat-panel.tsx b/packages/app/src/components/im/group-chat-panel.tsx
index 196a1079..18dd8d3b 100644
--- a/packages/app/src/components/im/group-chat-panel.tsx
+++ b/packages/app/src/components/im/group-chat-panel.tsx
@@ -16,10 +16,8 @@ export function GroupChatPanel(props: GroupChatPanelProps) {
const [agents, setAgents] = createSignal([])
const [loading, setLoading] = createSignal(true)
- const { connected, messages: realtimeMessages, agentStatuses, typingMembers, send } = useIMWebSocket(
- client,
- () => props.groupID,
- )
+ const { connected, messages: realtimeMessages, agentStatuses, agentProgress, typingMembers, send } =
+ useIMWebSocket(client, () => props.groupID)
// The single desktop user identity, as the server assigns it. Used to exclude
// our own echoed typing/read events from the UI.
@@ -111,7 +109,7 @@ export function GroupChatPanel(props: GroupChatPanelProps) {
-
+ 0}>
diff --git a/packages/app/src/components/im/message-item.tsx b/packages/app/src/components/im/message-item.tsx
index 8f23acdf..af532e10 100644
--- a/packages/app/src/components/im/message-item.tsx
+++ b/packages/app/src/components/im/message-item.tsx
@@ -1,11 +1,14 @@
import { Show } from "solid-js"
import { AgentStatusChip } from "./agent-status-chip"
+import { AgentReasoningCard } from "./agent-reasoning-card"
import { MessageMetadataCard } from "./message-metadata-card"
-import type { LocalMessage } from "@/hooks/use-im-websocket"
+import type { AgentProgressPart, LocalMessage } from "@/hooks/use-im-websocket"
interface MessageItemProps {
message: LocalMessage
agentStatus?: { agentID: string; status: string }
+ /** Live reasoning/tool/text parts for the turn this message triggered. */
+ agentProgress?: AgentProgressPart[]
}
export function MessageItem(props: MessageItemProps) {
@@ -40,6 +43,15 @@ export function MessageItem(props: MessageItemProps) {
+ 0}>
+
+
+
+
+
diff --git a/packages/app/src/components/im/message-list.tsx b/packages/app/src/components/im/message-list.tsx
index 59b7c443..4bd3e5d5 100644
--- a/packages/app/src/components/im/message-list.tsx
+++ b/packages/app/src/components/im/message-list.tsx
@@ -1,19 +1,24 @@
import { For, Show, createEffect } from "solid-js"
import { MessageItem } from "./message-item"
-import type { LocalMessage } from "@/hooks/use-im-websocket"
+import type { AgentProgressPart, LocalMessage } from "@/hooks/use-im-websocket"
interface MessageListProps {
messages: LocalMessage[]
agentStatuses: Map
+ agentProgress: Map
}
export function MessageList(props: MessageListProps) {
let containerRef: HTMLDivElement | undefined
let bottomRef: HTMLDivElement | undefined
- // Auto-scroll to bottom on new messages
+ // Auto-scroll to bottom on new messages AND as live reasoning streams in, so
+ // the agent's in-progress thinking stays in view without manual scrolling.
createEffect(() => {
- if (props.messages.length > 0 && bottomRef) {
+ // Track both signals so streaming progress updates also trigger a scroll.
+ const count = props.messages.length
+ props.agentProgress
+ if (count > 0 && bottomRef) {
bottomRef.scrollIntoView({ behavior: "smooth" })
}
})
@@ -33,6 +38,7 @@ export function MessageList(props: MessageListProps) {
)}
diff --git a/packages/app/src/components/im/types.ts b/packages/app/src/components/im/types.ts
index 8e59b67d..9366a73c 100644
--- a/packages/app/src/components/im/types.ts
+++ b/packages/app/src/components/im/types.ts
@@ -1,10 +1,10 @@
-export interface AgentDescriptor {
- id: string
- name: string
- displayName: string
- description?: string
- visible: boolean
-}
+// The IM agent descriptor is defined canonically in the core package
+// (`packages/core/src/im/mention-parser.ts`) and served verbatim by the HTTP
+// API (`AgentDescriptorResponse = AgentDescriptor`). Re-export the canonical
+// types here instead of hand-mirroring them, so the frontend can never drift
+// from the backend contract. `Trigger`, `AutonomyLevel` and `AgentLimits` are
+// the sub-schemas referenced by `AgentDescriptor`.
+export type { AgentDescriptor, AgentLimits, AutonomyLevel, Trigger } from "@deepagent-code/core/im/mention-parser"
export interface IMGroup {
id: string
diff --git a/packages/app/src/components/prompt-input/scenario-override.test.ts b/packages/app/src/components/prompt-input/scenario-override.test.ts
index f1879cfc..97fb3acf 100644
--- a/packages/app/src/components/prompt-input/scenario-override.test.ts
+++ b/packages/app/src/components/prompt-input/scenario-override.test.ts
@@ -12,50 +12,50 @@ import {
// choosing the turn's prompt pipeline mode.
describe("scenario override (D1/D3)", () => {
test("override is read back per key", () => {
- setScenarioOverride("ses_a", "wish")
- expect(getScenarioOverride("ses_a")).toBe("wish")
+ setScenarioOverride("ses_a", "intelligence")
+ expect(getScenarioOverride("ses_a")).toBe("intelligence")
setScenarioOverride("ses_a", "direct")
expect(getScenarioOverride("ses_a")).toBe("direct")
})
test("keys are isolated", () => {
- setScenarioOverride("ses_b", "wish")
+ setScenarioOverride("ses_b", "intelligence")
expect(getScenarioOverride("ses_c")).toBeUndefined()
})
test("resolve prefers session key, falls back to directory key", () => {
- setScenarioOverride("dir_x", "wish")
- // No session override yet -> directory-scoped wish applies (toggle set before session exists).
- expect(resolveScenarioOverride("ses_new", "dir_x")).toBe("wish")
+ setScenarioOverride("dir_x", "intelligence")
+ // No session override yet -> directory-scoped intelligence applies (toggle set before session exists).
+ expect(resolveScenarioOverride("ses_new", "dir_x")).toBe("intelligence")
// A session override takes precedence over the directory one.
setScenarioOverride("ses_new", "direct")
expect(resolveScenarioOverride("ses_new", "dir_x")).toBe("direct")
})
test("stop clears both session and directory overrides so the next turn uses the configured default", () => {
- setScenarioOverride("ses_d", "wish")
- setScenarioOverride("dir_d", "wish")
+ setScenarioOverride("ses_d", "intelligence")
+ setScenarioOverride("dir_d", "intelligence")
resetScenarioOnStop("ses_d", "dir_d")
// Cleared, not pinned to "direct": resolution now falls through to the configured default,
- // and a pinned session key can no longer shadow a later dir-key toggle back to wish.
+ // and a pinned session key can no longer shadow a later dir-key toggle back to intelligence.
expect(getScenarioOverride("ses_d")).toBeUndefined()
expect(getScenarioOverride("dir_d")).toBeUndefined()
expect(resolveScenarioOverride("ses_d", "dir_d")).toBeUndefined()
})
- test("after stop, re-toggling wish on the directory key re-engages wish at submit time", () => {
- // Repro of the 'exit wish -> can never re-enter' bug: previously stop pinned the SESSION key to
+ test("after stop, re-toggling intelligence on the directory key re-engages intelligence at submit time", () => {
+ // Repro of the 'exit intelligence -> can never re-enter' bug: previously stop pinned the SESSION key to
// "direct", which submit resolves before the dir key, so the toggle (dir key only) was shadowed.
- setScenarioOverride("ses_f", "wish")
+ setScenarioOverride("ses_f", "intelligence")
resetScenarioOnStop("ses_f", "dir_f")
- setScenarioOverride("dir_f", "wish") // user re-selects wish via the toggle
- expect(resolveScenarioOverride("ses_f", "dir_f")).toBe("wish")
+ setScenarioOverride("dir_f", "intelligence") // user re-selects intelligence via the toggle
+ expect(resolveScenarioOverride("ses_f", "dir_f")).toBe("intelligence")
})
test("listeners are notified when overrides change", () => {
let count = 0
const unsubscribe = subscribeScenarioOverride(() => count++)
- setScenarioOverride("ses_e", "wish")
+ setScenarioOverride("ses_e", "intelligence")
resetScenarioOnStop("ses_e")
unsubscribe()
setScenarioOverride("ses_e", "direct")
diff --git a/packages/app/src/components/prompt-input/scenario-override.ts b/packages/app/src/components/prompt-input/scenario-override.ts
index f098cb3f..58eae0f4 100644
--- a/packages/app/src/components/prompt-input/scenario-override.ts
+++ b/packages/app/src/components/prompt-input/scenario-override.ts
@@ -1,6 +1,6 @@
// D1/D3: scenario-mode override, set by the send-adjacent scenario toggle (D1).
// It overrides the configured promptMode for a turn. Cleared on stop (D3) so the override is
-// strictly per-turn: once a turn ends, resolution falls back to the configured default (e.g. wish)
+// strictly per-turn: once a turn ends, resolution falls back to the configured default (e.g. intelligence)
// rather than being pinned to a stale value. The next turn re-enters the configured scenario unless
// the user toggles again.
//
@@ -11,7 +11,7 @@
//
// This lives in its own module (no SolidJS imports) so submit.ts and the toggle UI can share it
// and it stays unit-testable without loading the router/runtime.
-export type ScenarioOverride = "direct" | "wish"
+export type ScenarioOverride = "direct" | "intelligence"
const scenarioOverride = new Map()
const listeners = new Set<() => void>()
@@ -39,7 +39,7 @@ export const resolveScenarioOverride = (sessionKey: string, dirKey: string): Sce
// D3: any stop CLEARS the per-turn override for both the session and its directory scope. Clearing
// (not pinning "direct") is deliberate: a pinned "direct" would permanently shadow the configured
// default and, because submit resolves the session key before the dir key, the toggle (which only
-// writes the dir key) could never re-engage wish. After clearing, the next turn falls back to the
+// writes the dir key) could never re-engage intelligence. After clearing, the next turn falls back to the
// configured scenario default.
export const resetScenarioOnStop = (sessionKey: string, dirKey?: string): void => {
scenarioOverride.delete(sessionKey)
diff --git a/packages/app/src/components/prompt-input/scenario-toggle.tsx b/packages/app/src/components/prompt-input/scenario-toggle.tsx
index ca942e0b..b7d25e2c 100644
--- a/packages/app/src/components/prompt-input/scenario-toggle.tsx
+++ b/packages/app/src/components/prompt-input/scenario-toggle.tsx
@@ -16,15 +16,15 @@ const scenarios = [
tooltip: "prompt.scenario.direct.tooltip" as const,
},
{
- mode: "wish" as const,
+ mode: "intelligence" as const,
icon: "speech-bubble" as const,
- label: "prompt.scenario.wish" as const,
- tooltip: "prompt.scenario.wish.tooltip" as const,
+ label: "prompt.scenario.intelligence" as const,
+ tooltip: "prompt.scenario.intelligence.tooltip" as const,
},
]
// D1: the per-turn scenario-mode toggle that sits to the left of the send button. It flips the
-// scenario between `direct` (the user owns the prompt) and `wish` (DeepAgent prepares the prompt
+// scenario between `direct` (the user owns the prompt) and `intelligence` (DeepAgent prepares the prompt
// and proposes next-round suggestions). It writes a DIRECTORY-scoped override (stable before a
// session exists) that submit.ts resolves session-then-directory, so a toggle made on the
// new-session composer still applies to the first turn. It defaults to the configured promptMode
@@ -39,12 +39,12 @@ export function ScenarioToggle() {
onCleanup(subscribeScenarioOverride(() => setVersion((value) => value + 1)))
// Effective scenario = directory override if set, else the configured default.
- const scenario = createMemo<"direct" | "wish">(() => {
+ const scenario = createMemo<"direct" | "intelligence">(() => {
version()
return getScenarioOverride(dirKey()) ?? deepAgentPromptModeFromConfig(serverSync.data.config)
})
- const select = (mode: "direct" | "wish") => setScenarioOverride(dirKey(), mode)
+ const select = (mode: "direct" | "intelligence") => setScenarioOverride(dirKey(), mode)
return (
{
})
if (text === "prepare fails") {
throw new Error("POST /session/ses_1/prompt_prepare returned 400", {
- cause: { body: { name: "BadRequest", data: { message: "Wish prompt preparation failed" } }, status: 400 },
+ cause: {
+ body: { name: "BadRequest", data: { message: "Intelligence prompt preparation failed" } },
+ status: 400,
+ },
})
}
return {
@@ -91,7 +96,7 @@ const clientFor = (directory: string) => {
prompt_draft_id: "prompt_draft:test:1",
context_plan_id: "context_plan:test:1",
state: "draft_ready",
- mode: payload.body?.mode ?? "wish",
+ mode: payload.body?.mode ?? "intelligence",
route: text === "hello" ? "general" : "code",
goal: "Prepared goal",
preview: "# Prepared prompt",
@@ -415,9 +420,9 @@ describe("prompt submit worktree selection", () => {
expect(optimisticSeeded).toEqual([true])
})
- test("prepares and confirms wish prompts before async submission", async () => {
+ test("prepares and confirms intelligence prompts before async submission", async () => {
params = { id: "session-1" }
- promptMode = "wish"
+ promptMode = "intelligence"
const submit = createPromptSubmit({
info: () => ({ id: "session-1" }),
@@ -446,13 +451,13 @@ describe("prompt submit worktree selection", () => {
expect(promptPrepareEvents).toEqual(["start", "end"])
expect(preparedDrafts).toEqual([
- { directory: "/repo/main", sessionID: "session-1", mode: "wish", outputLanguage: "english", text: "ls" },
+ { directory: "/repo/main", sessionID: "session-1", mode: "intelligence", outputLanguage: "english", text: "ls" },
])
expect(sentPromptAsync[0]?.text).toBe("Edited prepared goal")
expect(sentPromptAsync[0]?.metadata).toEqual({
deepagent: {
prompt_pipeline: {
- mode: "wish",
+ mode: "intelligence",
confirmed_draft_id: "prompt_draft:test:1",
edited_goal: "Edited prepared goal",
},
@@ -460,7 +465,9 @@ describe("prompt submit worktree selection", () => {
})
})
- test("prepares wish prompts in Chinese when the app locale is Chinese", async () => {
+ // Legacy-compat: a config still storing the pre-rename "wish" value must normalize to
+ // "intelligence" on the wire (read-old/send-new). Also exercises the Chinese output language.
+ test("prepares intelligence prompts in Chinese, normalizing a legacy 'wish' config to 'intelligence'", async () => {
params = { id: "session-1" }
promptMode = "wish"
appLocale = "zh"
@@ -487,13 +494,13 @@ describe("prompt submit worktree selection", () => {
await flushAsyncSubmit()
expect(preparedDrafts).toEqual([
- { directory: "/repo/main", sessionID: "session-1", mode: "wish", outputLanguage: "chinese", text: "ls" },
+ { directory: "/repo/main", sessionID: "session-1", mode: "intelligence", outputLanguage: "chinese", text: "ls" },
])
})
- test("routes non-code wish prompts through general without confirmation", async () => {
+ test("routes non-code intelligence prompts through general without confirmation", async () => {
params = { id: "session-1" }
- promptMode = "wish"
+ promptMode = "intelligence"
const confirms: string[] = []
promptValue[0] = { type: "text", content: "hello", start: 0, end: 5 }
@@ -525,7 +532,7 @@ describe("prompt submit worktree selection", () => {
expect(confirms).toEqual([])
expect(preparedDrafts).toEqual([
- { directory: "/repo/main", sessionID: "session-1", mode: "wish", outputLanguage: "english", text: "hello" },
+ { directory: "/repo/main", sessionID: "session-1", mode: "intelligence", outputLanguage: "english", text: "hello" },
])
expect(sentPromptAsync[0]?.text).toBe("hello")
expect(sentPromptAsync[0]?.metadata).toEqual({
@@ -539,9 +546,9 @@ describe("prompt submit worktree selection", () => {
promptValue[0] = { type: "text", content: "ls", start: 0, end: 2 }
})
- test("does not submit when wish prompt preparation fails", async () => {
+ test("does not submit when intelligence prompt preparation fails", async () => {
params = { id: "session-1" }
- promptMode = "wish"
+ promptMode = "intelligence"
promptValue[0] = { type: "text", content: "prepare fails", start: 0, end: 13 }
const submit = createPromptSubmit({
@@ -570,7 +577,7 @@ describe("prompt submit worktree selection", () => {
{
directory: "/repo/main",
sessionID: "session-1",
- mode: "wish",
+ mode: "intelligence",
outputLanguage: "english",
text: "prepare fails",
},
diff --git a/packages/app/src/components/prompt-input/submit.ts b/packages/app/src/components/prompt-input/submit.ts
index 24f350d3..baa67555 100644
--- a/packages/app/src/components/prompt-input/submit.ts
+++ b/packages/app/src/components/prompt-input/submit.ts
@@ -40,7 +40,7 @@ export type FollowupDraft = {
metadata?: {
deepagent?: {
prompt_pipeline?: {
- mode: "wish" | "direct_override"
+ mode: "intelligence" | "direct_override"
confirmed_draft_id?: string
edited_goal?: string
}
@@ -53,7 +53,7 @@ export type DeepAgentPromptPrepareResult = {
prompt_draft_id: string
context_plan_id: string
state: string
- mode: "wish"
+ mode: "intelligence"
route: "code" | "general"
goal: string
preview: string
@@ -102,7 +102,7 @@ const draftImages = (prompt: Prompt) => prompt.filter((part): part is ImageAttac
const promptPipelineMode = (metadata: FollowupDraft["metadata"]): DeepAgentPromptModeForConfirmation | undefined => {
const mode = metadata?.deepagent?.prompt_pipeline?.mode
- return mode === "wish" ? mode : undefined
+ return mode === "intelligence" ? mode : undefined
}
async function prepareDeepAgentPromptDraft(input: {
@@ -227,7 +227,7 @@ export async function sendFollowupDraft(input: FollowupSendInput) {
waited = true
if (!ok) return false
- // D2: while wish prepares the prompt (a real model call), surface a busy/"advice generating"
+ // D2: while intelligence prepares the prompt (a real model call), surface a busy/"advice generating"
// state so the composer indicates work is in progress rather than appearing frozen. The
// existing confirm surface then shows the prepared prompt for review/edit before send.
setBusy()
@@ -397,13 +397,9 @@ export function createPromptSubmit(input: PromptSubmitInput) {
// D3: any stop resets the scenario to `direct` and pauses scenario automation for this
// session and its directory scope, so the next user message is a plain direct submission
- // until they re-engage wish.
+ // until they re-engage intelligence.
resetScenarioOnStop(pendingKey(sessionID), ScopedKey.from(sdk.scope, sdk.directory))
- serverSync.todo.set(sessionID, [])
- const [, setStore] = serverSync.child(sdk.directory)
- setStore("todo", sessionID, [])
-
input.onAbort?.()
const key = pendingKey(sessionID)
@@ -585,11 +581,11 @@ export function createPromptSubmit(input: PromptSubmitInput) {
// selection swaps `sessionDirectory` to the new worktree dir). Read the same project-scoped
// key so the toggle's override is found at submit time even in the worktree case; resolve
// session-then-directory. After a stop (D3) the override is "direct", so the next turn is a
- // plain direct submission until the user re-engages wish.
+ // plain direct submission until the user re-engages intelligence.
const promptMode =
resolveScenarioOverride(pendingKey(session.id), ScopedKey.from(sdk.scope, projectDirectory)) ??
deepAgentPromptModeFromConfig(serverSync.data.config)
- if (promptMode === "wish") {
+ if (promptMode === "intelligence") {
draft.metadata = {
deepagent: {
prompt_pipeline: {
@@ -681,7 +677,7 @@ export function createPromptSubmit(input: PromptSubmitInput) {
const commentItems = context.filter((item) => item.type === "file" && !!item.comment?.trim())
const messageID = Identifier.ascending("message")
- const preparesPromptDraft = promptPipelineMode(draft.metadata) === "wish"
+ const preparesPromptDraft = promptPipelineMode(draft.metadata) === "intelligence"
const removeOptimisticMessage = () => {
sync.session.optimistic.remove({
diff --git a/packages/app/src/components/session/session-header.tsx b/packages/app/src/components/session/session-header.tsx
index ed1c71da..c3cbe55a 100644
--- a/packages/app/src/components/session/session-header.tsx
+++ b/packages/app/src/components/session/session-header.tsx
@@ -245,6 +245,13 @@ export function SessionHeader() {
const tint = createMemo(() =>
messageAgentColor(params.id ? sync.data.message[params.id] : undefined, sync.data.agent),
)
+ // A running subagent (child session, parentID === current) puts a pulsing dot on the side-panel
+ // toggle so the user sees a spawn happened without opening the Subagents panel first.
+ const hasRunningSubagent = createMemo(() => {
+ const id = params.id
+ if (!id) return false
+ return sync.data.session.some((s) => s.parentID === id && sync.data.session_working(s.id))
+ })
const selectApp = (app: OpenApp) => {
if (!options().some((item) => item.id === app)) return
setPrefs("app", app)
@@ -463,13 +470,20 @@ export function SessionHeader() {
>
diff --git a/packages/app/src/components/settings-v2/general.tsx b/packages/app/src/components/settings-v2/general.tsx
index e6ad5a48..2c076f06 100644
--- a/packages/app/src/components/settings-v2/general.tsx
+++ b/packages/app/src/components/settings-v2/general.tsx
@@ -35,11 +35,13 @@ import "./settings-v2.css"
import {
deepAgentModeFromConfig,
deepAgentPromptModeFromConfig,
- deepAgentWishModelFromConfig,
+ deepAgentIntelligenceModelFromConfig,
deepAgentSelfLearningFromConfig,
+ deepAgentSubagentIntensityFromConfig,
type DeepAgentMode,
type DeepAgentPromptMode,
type DeepAgentSelfLearning,
+ type DeepAgentSubagentIntensity,
updateDeepAgentOptions,
} from "@/utils/deepagent-settings"
import { useModels } from "@/context/models"
@@ -205,15 +207,15 @@ export const SettingsGeneralV2: Component = () => {
description: language.t("settings.general.deepagent.prompt.direct.description"),
},
{
- value: "wish",
- label: language.t("settings.general.deepagent.prompt.wish"),
- description: language.t("settings.general.deepagent.prompt.wish.description"),
+ value: "intelligence",
+ label: language.t("settings.general.deepagent.prompt.intelligence"),
+ description: language.t("settings.general.deepagent.prompt.intelligence.description"),
},
])
const scenarioModeDescription = createMemo(
() => scenarioModeCards().find((card) => card.value === deepAgentPromptMode())?.description ?? "",
)
- // #3: `ultra` requires the wish scenario; it cannot run under `direct`. Gate against the
+ // #3: `ultra` requires the intelligence scenario; it cannot run under `direct`. Gate against the
// config-level scenario mode (this selector is a session/config-level setting, not per-turn).
// When the scenario is `direct`, drop `ultra` from the selectable options (unless it is somehow
// already the current value, so the control still reflects state) and surface an explanatory
@@ -231,8 +233,15 @@ export const SettingsGeneralV2: Component = () => {
}
return options
})
- const deepAgentWishModel = createMemo(() => deepAgentWishModelFromConfig(serverSync.data.config) ?? "")
- const wishModelOptions = createMemo(() =>
+ const deepAgentSubagentIntensity = createMemo(() => deepAgentSubagentIntensityFromConfig(serverSync.data.config))
+ const subagentIntensityOptions = createMemo<{ value: DeepAgentSubagentIntensity; label: string }[]>(() => [
+ { value: "inherit", label: language.t("settings.general.deepagent.subagentMode.inherit") },
+ { value: "downgrade", label: language.t("settings.general.deepagent.subagentMode.downgrade") },
+ ])
+ const deepAgentIntelligenceModel = createMemo(
+ () => deepAgentIntelligenceModelFromConfig(serverSync.data.config) ?? "",
+ )
+ const intelligenceModelOptions = createMemo(() =>
models.list().map((model) => ({
value: `${model.provider.id}/${model.id}`,
label: model.name,
@@ -357,7 +366,9 @@ export const SettingsGeneralV2: Component = () => {
{language.t("settings.general.deepagent.mode.description")}
{" "}
- {language.t("settings.general.deepagent.mode.ultraRequiresWish")}
+
+ {language.t("settings.general.deepagent.mode.ultraRequiresIntelligence")}
+
}
@@ -380,6 +391,27 @@ export const SettingsGeneralV2: Component = () => {
/>
+
+ o.value === deepAgentSubagentIntensity())}
+ value={(o) => o.value}
+ label={(o) => o.label}
+ onSelect={(option) => {
+ if (!option) return
+ if (option.value === deepAgentSubagentIntensity()) return
+ void updateDeepAgentOptions(serverSync, { subagentIntensity: option.value })
+ }}
+ />
+
+
{
option.value === deepAgentWishModel()) ?? wishModelOptions()[0]
+ intelligenceModelOptions().find((option) => option.value === deepAgentIntelligenceModel()) ??
+ intelligenceModelOptions()[0]
}
value={(o) => o.value}
label={(o) => o.label}
onSelect={(option) => {
if (!option) return
- if (option.value === deepAgentWishModel()) return
- void updateDeepAgentOptions(serverSync, { wishModel: option.value })
+ if (option.value === deepAgentIntelligenceModel()) return
+ void updateDeepAgentOptions(serverSync, { intelligenceModel: option.value })
}}
>
{(o) => (
diff --git a/packages/app/src/context/directory-sync.ts b/packages/app/src/context/directory-sync.ts
index c2957286..7663c2cf 100644
--- a/packages/app/src/context/directory-sync.ts
+++ b/packages/app/src/context/directory-sync.ts
@@ -191,7 +191,6 @@ export const createDirSyncContext = (
const historyMessagePageSize = 200
const inflight = new Map>()
const inflightDiff = new Map>()
- const inflightTodo = new Map>()
const optimistic = new Map>()
const maxDirs = 30
const seen = new Map>()
@@ -277,9 +276,6 @@ export const createDirSyncContext = (
const evict = (directory: string, setStore: Setter, sessionIDs: string[]) => {
if (sessionIDs.length === 0) return
clearSessionPrefetch(serverSDK.scope, directory, sessionIDs)
- for (const sessionID of sessionIDs) {
- serverSync.todo.set(sessionID, undefined)
- }
setStore(
produce((draft) => {
dropSessionCaches(draft, sessionIDs)
@@ -522,32 +518,6 @@ export const createDirSyncContext = (
}),
)
},
- async todo(sessionID: string, opts?: { force?: boolean }) {
- const [store, setStore] = serverSync.child(directory)
- touch(directory, setStore, sessionID)
- const existing = store.todo[sessionID]
- const cached = serverSync.data.session_todo[sessionID]
- if (existing !== undefined) {
- if (cached === undefined) {
- serverSync.todo.set(sessionID, existing)
- }
- if (!opts?.force) return
- }
-
- if (cached !== undefined) {
- setStore("todo", sessionID, reconcile(cached, { key: "id" }))
- }
-
- const key = keyFor(directory, sessionID)
- return runInflight(inflightTodo, key, () =>
- retry(() => client.session.todo({ sessionID })).then((todo) => {
- if (!tracked(directory, sessionID)) return
- const list = todo.data ?? []
- setStore("todo", sessionID, reconcile(list, { key: "id" }))
- serverSync.todo.set(sessionID, list)
- }),
- )
- },
history: {
more(sessionID: string) {
const store = current()[0]
diff --git a/packages/app/src/context/global-sync/bootstrap.ts b/packages/app/src/context/global-sync/bootstrap.ts
index 343f2880..fe66bb35 100644
--- a/packages/app/src/context/global-sync/bootstrap.ts
+++ b/packages/app/src/context/global-sync/bootstrap.ts
@@ -7,7 +7,6 @@ import type {
ProviderAuthResponse,
QuestionRequest,
Session,
- Todo,
} from "@deepagent-code/sdk/v2/client"
import { showToast } from "@/utils/toast"
import { getFilename } from "@deepagent-code/core/util/path"
@@ -26,9 +25,6 @@ type GlobalStore = {
ready: boolean
path: Path
project: Project[]
- session_todo: {
- [sessionID: string]: Todo[]
- }
session_plan: {
[sessionID: string]: import("./types").SessionPlan
}
diff --git a/packages/app/src/context/global-sync/event-reducer.test.ts b/packages/app/src/context/global-sync/event-reducer.test.ts
index 05cdd503..29202b08 100644
--- a/packages/app/src/context/global-sync/event-reducer.test.ts
+++ b/packages/app/src/context/global-sync/event-reducer.test.ts
@@ -1,7 +1,9 @@
import { describe, expect, test } from "bun:test"
import type { Message, Part, PermissionRequest, Project, QuestionRequest, Session } from "@deepagent-code/sdk/v2/client"
-import { createStore } from "solid-js/store"
-import type { State } from "./types"
+import { createRoot } from "solid-js"
+import { isServer } from "solid-js/web"
+import { createStore, reconcile, unwrap } from "solid-js/store"
+import type { SessionPlan, State } from "./types"
import { applyDirectoryEvent, applyGlobalEvent, cleanupDroppedSessionCaches } from "./event-reducer"
const rootSession = (input: { id: string; parentID?: string; archived?: number }) =>
@@ -274,7 +276,6 @@ describe("applyDirectoryEvent", () => {
const dropped = rootSession({ id: "ses_b" })
const kept = rootSession({ id: "ses_a" })
const message = userMessage("msg_1", dropped.id)
- const todos: string[] = []
const [store, setStore] = createStore(
baseState({
limit: 1,
@@ -296,10 +297,6 @@ describe("applyDirectoryEvent", () => {
push() {},
directory: "/tmp",
loadLsp() {},
- setSessionTodo(sessionID, value) {
- if (value !== undefined) return
- todos.push(sessionID)
- },
})
expect(store.session.map((x) => x.id)).toEqual([kept.id])
@@ -310,7 +307,6 @@ describe("applyDirectoryEvent", () => {
expect(store.permission[dropped.id]).toBeUndefined()
expect(store.question[dropped.id]).toBeUndefined()
expect(store.session_status[dropped.id]).toBeUndefined()
- expect(todos).toEqual([dropped.id])
})
test("cleanupDroppedSessionCaches clears part-only orphan state", () => {
@@ -574,3 +570,125 @@ describe("applyDirectoryEvent", () => {
expect(lspLoads).toBe(1)
})
})
+
+// Regression guard for the plan panel. The model pushes repeated plan.updated events that advance
+// step statuses; the reducer feeds them into a per-session plan store via reconcile. The previous
+// code used `{ key: "plan_id" }`, which reconcile applies recursively to the nested `steps[]` array
+// — but a step's identity field is `step_id`, not `plan_id`, so every step resolved to
+// `key=undefined`.
+//
+// IMPORTANT (verified against solid-js@1.9.10): the wrong key does NOT drop status updates — field
+// values land correctly under plan_id, step_id, or null alike (see the status-advance test below,
+// which passes under all three keys). The only behavioural difference is per-step PROXY IDENTITY on
+// reorder, and even that is invisible to the current UI: the dock renders via (positional)
+// over plain objects the `planAsTodos` memo re-creates every tick, so store-proxy identity is never
+// consumed by the render. `key: "step_id"` is therefore the correct minimal-diff choice on data-
+// contract grounds (field-level updates instead of whole-object replace, correct identity if the
+// render ever moves to a keyed ), not a fix for a reproducible "stuck" symptom at this layer.
+// The status-advance test IS the meaningful CI regression guard; the identity test below only
+// documents the proxy-identity contract and requires --conditions=browser (see its comment).
+describe("plan.updated reconcile (session_plan)", () => {
+ const planEvent = (
+ sessionID: string,
+ steps: Array<[step_id: string, status: string]>,
+ activeStepID: string | null,
+ ) => ({
+ type: "plan.updated",
+ properties: {
+ sessionID,
+ plan_id: "plan_1",
+ goal: "ship it",
+ active_step_id: activeStepID,
+ steps: steps.map(([step_id, status]) => ({ step_id, title: step_id.toUpperCase(), status })),
+ done: steps.filter(([, status]) => status === "done").length,
+ total: steps.length,
+ },
+ })
+
+ // Mirror the real setSessionPlan writer in server-sync.tsx: session_plan[sid] = reconcile(plan, ...).
+ const makeSetSessionPlan = (
+ setPlanStore: (path: "sp", sid: string, value: unknown) => void,
+ key: string | null,
+ ) => {
+ return (sessionID: string, plan: SessionPlan | undefined) => {
+ if (!plan) return
+ setPlanStore("sp", sessionID, reconcile(plan, { key }) as never)
+ }
+ }
+
+ const dispatch = (setStore: any, store: any, setSessionPlan: any, event: any) => {
+ applyDirectoryEvent({
+ event,
+ store,
+ setStore,
+ push() {},
+ directory: "/tmp",
+ loadLsp() {},
+ setSessionPlan,
+ })
+ }
+
+ test("two consecutive plan.updated events advance step status through the reducer", () => {
+ createRoot((dispose) => {
+ const [store, setStore] = createStore(baseState())
+ const [planStore, setPlanStore] = createStore<{ sp: Record }>({ sp: {} })
+ const setSessionPlan = makeSetSessionPlan(setPlanStore as never, "step_id")
+
+ // First plan.updated: step "a" is active, the rest pending.
+ dispatch(setStore, store, setSessionPlan, planEvent("ses_1", [["a", "active"], ["b", "pending"], ["c", "pending"]], "a"))
+ expect(planStore.sp.ses_1.steps.map((s) => s.status)).toEqual(["active", "pending", "pending"])
+
+ // Second plan.updated: same plan_id, "a" done and "b" now active. This is the event the old
+ // code effectively dropped from the panel's point of view.
+ dispatch(setStore, store, setSessionPlan, planEvent("ses_1", [["a", "done"], ["b", "active"], ["c", "pending"]], "b"))
+ expect(planStore.sp.ses_1.steps.map((s) => s.status)).toEqual(["done", "active", "pending"])
+
+ // Third advance, to be thorough.
+ dispatch(setStore, store, setSessionPlan, planEvent("ses_1", [["a", "done"], ["b", "done"], ["c", "active"]], "c"))
+ expect(planStore.sp.ses_1.steps.map((s) => s.status)).toEqual(["done", "done", "active"])
+ expect(planStore.sp.ses_1.active_step_id).toBe("c")
+ expect(planStore.sp.ses_1.done).toBe(2)
+
+ dispose()
+ })
+ })
+
+ // Per-step proxy identity across reorder only manifests under the CLIENT (browser) build of
+ // Solid's store; the SSR build never retains proxy identity, so this assertion is only meaningful
+ // under `--conditions=browser`. CI GAP: the package `test`/`test:ci` scripts run SSR-only, so this
+ // test is skipped in CI today. It is NOT a user-visible regression guard — the dock renders steps
+ // positionally via over memo-recreated plain objects, so proxy identity never reaches the
+ // render. It documents the reconcile identity contract for a future keyed- render only. To
+ // exercise it locally: `bun test --conditions=browser --preload ./happydom.ts `.
+ test.skipIf(isServer)("keying by step_id preserves per-step identity across reorder", () => {
+ const mk = (rows: Array<[string, string]>): SessionPlan => ({
+ plan_id: "plan_1",
+ goal: "g",
+ active_step_id: rows[0]?.[0] ?? null,
+ steps: rows.map(([step_id, status]) => ({ step_id, title: step_id.toUpperCase(), status })),
+ done: rows.filter(([, s]) => s === "done").length,
+ total: rows.length,
+ })
+
+ const identityStable = (key: string) =>
+ createRoot((dispose) => {
+ const [store, setStore] = createStore<{ p?: SessionPlan }>({})
+ setStore("p", reconcile(mk([["a", "active"], ["b", "pending"]]), { key }))
+ const before = unwrap(store.p!.steps.find((s) => s.step_id === "a")!)
+ setStore("p", reconcile(mk([["b", "active"], ["a", "done"]]), { key }))
+ const after = unwrap(store.p!.steps.find((s) => s.step_id === "a")!)
+ dispose()
+ return before === after
+ })
+
+ // The fix keys by step_id and keeps a step's identity stable when the array reorders.
+ expect(identityStable("step_id")).toBe(true)
+ // The old (buggy) key does not.
+ expect(identityStable("plan_id")).toBe(false)
+ })
+})
+
+// NOTE: the `todo.updated reconcile` describe block was removed here. Task tracking is unified onto
+// the plan system: the backend no longer emits `todo.updated` (both todowrite tool writers were
+// removed) and the reducer no longer handles it. The plan panel's live-update coverage lives in the
+// `plan.updated reconcile (session_plan)` describe block below.
diff --git a/packages/app/src/context/global-sync/event-reducer.ts b/packages/app/src/context/global-sync/event-reducer.ts
index f57782dd..b2cd5cb2 100644
--- a/packages/app/src/context/global-sync/event-reducer.ts
+++ b/packages/app/src/context/global-sync/event-reducer.ts
@@ -9,7 +9,6 @@ import type {
Session,
SessionStatus,
SnapshotFileDiff,
- Todo,
} from "@deepagent-code/sdk/v2/client"
import type { State, VcsCache, SessionPlan } from "./types"
import { trimSessions } from "./session-trim"
@@ -47,13 +46,8 @@ export function applyGlobalEvent(input: {
)
}
-function cleanupSessionCaches(
- setStore: SetStoreFunction,
- sessionID: string,
- setSessionTodo?: (sessionID: string, todos: Todo[] | undefined) => void,
-) {
+function cleanupSessionCaches(setStore: SetStoreFunction, sessionID: string) {
if (!sessionID) return
- setSessionTodo?.(sessionID, undefined)
setStore(
produce((draft) => {
dropSessionCaches(draft, [sessionID])
@@ -61,12 +55,7 @@ function cleanupSessionCaches(
)
}
-export function cleanupDroppedSessionCaches(
- store: Store,
- setStore: SetStoreFunction,
- next: Session[],
- setSessionTodo?: (sessionID: string, todos: Todo[] | undefined) => void,
-) {
+export function cleanupDroppedSessionCaches(store: Store, setStore: SetStoreFunction, next: Session[]) {
const keep = new Set(next.map((item) => item.id))
const stale = [
...Object.keys(store.message),
@@ -80,9 +69,6 @@ export function cleanupDroppedSessionCaches(
.filter((sessionID): sessionID is string => !!sessionID),
].filter((sessionID, index, list) => !keep.has(sessionID) && list.indexOf(sessionID) === index)
if (stale.length === 0) return
- for (const sessionID of stale) {
- setSessionTodo?.(sessionID, undefined)
- }
setStore(
produce((draft) => {
dropSessionCaches(draft, stale)
@@ -98,7 +84,6 @@ export function applyDirectoryEvent(input: {
directory: string
loadLsp: () => void
vcsCache?: VcsCache
- setSessionTodo?: (sessionID: string, todos: Todo[] | undefined) => void
setSessionPlan?: (sessionID: string, plan: SessionPlan | undefined) => void
retainedLimit?: number
}) {
@@ -120,7 +105,7 @@ export function applyDirectoryEvent(input: {
next.splice(result.index, 0, info)
const trimmed = trimSessions(next, { limit, permission: input.store.permission })
input.setStore("session", reconcile(trimmed, { key: "id" }))
- cleanupDroppedSessionCaches(input.store, input.setStore, trimmed, input.setSessionTodo)
+ cleanupDroppedSessionCaches(input.store, input.setStore, trimmed)
if (!info.parentID) input.setStore("sessionTotal", (value) => value + 1)
break
}
@@ -137,7 +122,7 @@ export function applyDirectoryEvent(input: {
}),
)
}
- cleanupSessionCaches(input.setStore, info.id, input.setSessionTodo)
+ cleanupSessionCaches(input.setStore, info.id)
if (info.parentID) break
input.setStore("sessionTotal", (value) => Math.max(0, value - 1))
break
@@ -150,7 +135,7 @@ export function applyDirectoryEvent(input: {
next.splice(result.index, 0, info)
const trimmed = trimSessions(next, { limit, permission: input.store.permission })
input.setStore("session", reconcile(trimmed, { key: "id" }))
- cleanupDroppedSessionCaches(input.store, input.setStore, trimmed, input.setSessionTodo)
+ cleanupDroppedSessionCaches(input.store, input.setStore, trimmed)
break
}
case "session.deleted": {
@@ -164,7 +149,7 @@ export function applyDirectoryEvent(input: {
}),
)
}
- cleanupSessionCaches(input.setStore, info.id, input.setSessionTodo)
+ cleanupSessionCaches(input.setStore, info.id)
if (info.parentID) break
input.setStore("sessionTotal", (value) => Math.max(0, value - 1))
break
@@ -174,16 +159,12 @@ export function applyDirectoryEvent(input: {
input.setStore("session_diff", props.sessionID, reconcile(list(props.diff), { key: "file" }))
break
}
- case "todo.updated": {
- const props = event.properties as { sessionID: string; todos: Todo[] }
- input.setStore("todo", props.sessionID, reconcile(props.todos, { key: "id" }))
- input.setSessionTodo?.(props.sessionID, props.todos)
- break
- }
+ // NOTE: the `todo.updated` event handler was removed when task tracking unified onto the plan
+ // system. The backend no longer emits `todo.updated` (both todowrite tool writers were removed),
+ // and the dock renders the plan exclusively. See `plan.updated` below for the live task source.
case "plan.updated": {
- // U2: the live plan (goal + steps + progress) from the `plan` tool. Stored under a distinct
- // session_plan key so it persists when the session goes idle (unlike the todo cache, which the
- // composer clears on idle).
+ // The live plan (goal + steps + progress) from the `plan` tool. Stored under a distinct
+ // session_plan key so it persists when the session goes idle.
const props = event.properties as {
sessionID: string
plan_id: string
diff --git a/packages/app/src/context/server-sync.tsx b/packages/app/src/context/server-sync.tsx
index 74ebb090..83133d3c 100644
--- a/packages/app/src/context/server-sync.tsx
+++ b/packages/app/src/context/server-sync.tsx
@@ -6,7 +6,6 @@ import type {
Path,
Project,
ProviderAuthResponse,
- Todo,
} from "@deepagent-code/sdk/v2/client"
import { showToast } from "@/utils/toast"
import { getFilename } from "@deepagent-code/core/util/path"
@@ -55,11 +54,9 @@ type GlobalStore = {
error?: InitError
path: Path
project: Project[]
- session_todo: {
- [sessionID: string]: Todo[]
- }
- // U2: the live plan per session (goal + steps + progress) pushed by the `plan` tool's
- // plan.updated event. Persistent — survives a session going idle (unlike the todo cache).
+ // The live plan per session (goal + steps + progress) pushed by the `plan` tool's plan.updated
+ // event. Persistent — survives a session going idle. This is the SINGLE source for the task dock;
+ // the legacy `session_todo` cache was removed when task tracking unified onto the plan system.
session_plan: {
[sessionID: string]: SessionPlan
}
@@ -135,7 +132,6 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) {
return !bootstrap.isPending
},
project: [],
- session_todo: {},
session_plan: {},
provider_auth: {},
get path() {
@@ -238,21 +234,7 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) {
return (setGlobalStore as (...args: unknown[]) => unknown)(...input)
}) as typeof setGlobalStore
- const setSessionTodo = (sessionID: string, todos: Todo[] | undefined) => {
- if (!sessionID) return
- if (!todos) {
- setGlobalStore(
- "session_todo",
- produce((draft) => {
- delete draft[sessionID]
- }),
- )
- return
- }
- setGlobalStore("session_todo", sessionID, reconcile(todos, { key: "id" }))
- }
-
- // U2: set/clear the live plan for a session.
+ // Set/clear the live plan for a session.
const setSessionPlan = (sessionID: string, plan: SessionPlan | undefined) => {
if (!sessionID) return
if (!plan) {
@@ -264,7 +246,16 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) {
)
return
}
- setGlobalStore("session_plan", sessionID, reconcile(plan, { key: "plan_id" }))
+ // The outer object is a single plan per session (not an array), so a key is meaningless at this
+ // level — but reconcile applies `key` RECURSIVELY to the nested `steps[]` array, where the
+ // identity field is `step_id`, not `plan_id`. With the wrong key every step resolves to
+ // `key=undefined`, so reconcile can't match steps across updates and falls back to replacing
+ // whole step objects by position (extra store-cell churn). NOTE: this does NOT drop status
+ // changes — field values still land either way (verified), and the dock renders steps via
+ // over plain objects re-created by the `planAsTodos` memo, so proxy identity is never
+ // consumed by the render. Keying by `step_id` is the correct, minimal-diff choice (field-level
+ // updates, stable identity) and is future-proof if the render ever switches to a keyed .
+ setGlobalStore("session_plan", sessionID, reconcile(plan, { key: "step_id" }))
}
const paused = () => untrack(() => globalStore.reload) !== undefined
@@ -332,7 +323,7 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) {
})
if (next.length !== store.session.length) {
setStore("session", reconcile(next, { key: "id" }))
- cleanupDroppedSessionCaches(store, setStore, next, setSessionTodo)
+ cleanupDroppedSessionCaches(store, setStore, next)
}
children.unpin(key)
return
@@ -369,7 +360,7 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) {
}),
)
setStore("session", reconcile(sessions, { key: "id" }))
- cleanupDroppedSessionCaches(store, setStore, sessions, setSessionTodo)
+ cleanupDroppedSessionCaches(store, setStore, sessions)
})
sessionMeta.set(key, { limit })
})
@@ -469,7 +460,6 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) {
store,
setStore,
push: queue.push,
- setSessionTodo,
setSessionPlan,
retainedLimit: sessionMeta.get(key)?.limit,
vcsCache: children.vcsCache.get(key),
@@ -618,9 +608,6 @@ export function createServerSyncContextInner(_serverSDK?: ServerSDK) {
updateConfig: updateConfigMutation.mutateAsync,
refreshProviders,
project: projectApi,
- todo: {
- set: setSessionTodo,
- },
plan: {
set: setSessionPlan,
},
diff --git a/packages/app/src/hooks/use-im-websocket.ts b/packages/app/src/hooks/use-im-websocket.ts
index 55aca6eb..0fc15337 100644
--- a/packages/app/src/hooks/use-im-websocket.ts
+++ b/packages/app/src/hooks/use-im-websocket.ts
@@ -2,11 +2,25 @@ import { createSignal, createEffect, onCleanup } from "solid-js"
import type { IMMessage } from "@/components/im/types"
import type { IMClient } from "@/utils/im-client"
+// One live snapshot of a part of an in-flight agent turn (reasoning / assistant
+// text / tool activity). The client keeps a map keyed by `partID` and REPLACES
+// each entry as batches arrive, so a dropped/reordered batch self-heals on the
+// next snapshot. `order` gives stable render order.
+export interface AgentProgressPart {
+ partID: string
+ order: number
+ kind: "reasoning" | "text" | "tool"
+ text?: string
+ tool?: string
+ status?: string
+}
+
// WebSocket event types
export type IMWebSocketEvent =
| { type: "message_created"; data: IMMessage }
| { type: "message_failed"; data: { clientMessageID?: string; code: string; message: string; retryable: boolean } }
| { type: "agent_status"; data: { messageID: string; agentID: string; status: string; error?: any } }
+ | { type: "agent_progress"; data: { messageID: string; agentID: string; parts: AgentProgressPart[] } }
| { type: "typing"; data: { groupID: string; memberID: string; typing: boolean } }
| { type: "read_receipt"; data: { groupID: string; memberID: string; readAt: number } }
| { type: "ping"; data: { ts: number } }
@@ -31,6 +45,9 @@ export function useIMWebSocket(client: IMClient, groupID: () => string | null) {
const [reconnectAttempt, setReconnectAttempt] = createSignal(0)
const [messages, setMessages] = createSignal([])
const [agentStatuses, setAgentStatuses] = createSignal>(new Map())
+ // trigger messageID -> ordered live reasoning/tool/text parts for that turn.
+ // Parts are keyed by partID within each turn and REPLACED as batches arrive.
+ const [agentProgress, setAgentProgress] = createSignal>(new Map())
// memberID -> last read timestamp, populated from read_receipt events.
const [readReceipts, setReadReceipts] = createSignal>(new Map())
// memberIDs currently typing (excluding self, resolved by the caller).
@@ -98,6 +115,21 @@ export function useIMWebSocket(client: IMClient, groupID: () => string | null) {
})
break
+ case "agent_progress":
+ // Merge the batch into this turn's parts: replace by partID (a part
+ // may update many times), then re-sort by `order` so the reasoning
+ // card renders in production order regardless of batch arrival.
+ setAgentProgress((prev) => {
+ const next = new Map(prev)
+ const existing = next.get(wsEvent.data.messageID) ?? []
+ const byPart = new Map(existing.map((p) => [p.partID, p]))
+ for (const part of wsEvent.data.parts) byPart.set(part.partID, part)
+ const merged = Array.from(byPart.values()).sort((a, b) => a.order - b.order)
+ next.set(wsEvent.data.messageID, merged)
+ return next
+ })
+ break
+
case "ping":
// Respond to ping
if (socket.readyState === WebSocket.OPEN) {
@@ -173,6 +205,7 @@ export function useIMWebSocket(client: IMClient, groupID: () => string | null) {
setMessages([])
setAgentStatuses(new Map())
+ setAgentProgress(new Map())
setReadReceipts(new Map())
setTypingMembers(new Set())
for (const timer of typingTimers.values()) clearTimeout(timer)
@@ -207,6 +240,7 @@ export function useIMWebSocket(client: IMClient, groupID: () => string | null) {
connected,
messages,
agentStatuses,
+ agentProgress,
readReceipts,
typingMembers,
send,
diff --git a/packages/app/src/i18n/ar.ts b/packages/app/src/i18n/ar.ts
index 09cf9968..ab9107c6 100644
--- a/packages/app/src/i18n/ar.ts
+++ b/packages/app/src/i18n/ar.ts
@@ -856,15 +856,19 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "يتطلب ultra وضع سيناريو wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "يتطلب ultra وضع سيناريو intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "وضع السيناريو",
"settings.general.deepagent.prompt.description": "اختر كيف يعالج المحرر الطلب قبل الإرسال.",
"settings.general.deepagent.prompt.direct": "الوضع المباشر",
"settings.general.deepagent.prompt.direct.description": "أرسل الطلب كما كتبته بدون أن يحضر DeepAgent مسودة.",
- "settings.general.deepagent.prompt.wish": "وضع wish",
- "settings.general.deepagent.prompt.wish.description": "يحضر DeepAgent طلبا كاملا أولا، ثم يتيح لك مراجعته وتعديله.",
- "settings.general.deepagent.wishModel.title": "نموذج وضع wish",
- "settings.general.deepagent.wishModel.description": "اختر النموذج المستخدم فقط لتحضير مسودات الطلبات في وضع wish.",
+ "settings.general.deepagent.prompt.intelligence": "وضع intelligence",
+ "settings.general.deepagent.prompt.intelligence.description": "يحضر DeepAgent طلبا كاملا أولا، ثم يتيح لك مراجعته وتعديله.",
+ "settings.general.deepagent.intelligenceModel.title": "نموذج وضع intelligence",
+ "settings.general.deepagent.intelligenceModel.description": "اختر النموذج المستخدم فقط لتحضير مسودات الطلبات في وضع intelligence.",
"settings.general.deepagent.selfLearning.title": "اعتماد التعلم الذاتي",
"settings.general.deepagent.selfLearning.description":
"حدد كيف تصبح المعرفة التي يتعلمها DeepAgent من جلساتك قابلة للاسترجاع.",
diff --git a/packages/app/src/i18n/br.ts b/packages/app/src/i18n/br.ts
index 26a1182e..95f636d3 100644
--- a/packages/app/src/i18n/br.ts
+++ b/packages/app/src/i18n/br.ts
@@ -870,18 +870,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra requiere el modo de escenario wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra requiere el modo de escenario intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Modo de cenário",
"settings.general.deepagent.prompt.description": "Elige cómo el compositor gestiona tu prompt antes de enviarlo.",
"settings.general.deepagent.prompt.direct": "Modo direto",
"settings.general.deepagent.prompt.direct.description":
"Envía tu prompt tal como está escrito, sin que DeepAgent prepare un borrador.",
- "settings.general.deepagent.prompt.wish": "Modo wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Modo intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent prepara primero un prompt completo y luego te permite revisarlo y editarlo.",
- "settings.general.deepagent.wishModel.title": "Modelo do modo wish",
- "settings.general.deepagent.wishModel.description":
- "Elige el modelo usado solo para preparar borradores de prompts del modo wish.",
+ "settings.general.deepagent.intelligenceModel.title": "Modelo do modo intelligence",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Elige el modelo usado solo para preparar borradores de prompts del modo intelligence.",
"settings.general.deepagent.selfLearning.title": "Aprobación de autoaprendizaje",
"settings.general.deepagent.selfLearning.description":
"Decide cómo el conocimiento que DeepAgent aprende de tus sesiones se vuelve recuperable.",
diff --git a/packages/app/src/i18n/bs.ts b/packages/app/src/i18n/bs.ts
index 72e1d0ca..72217dfa 100644
--- a/packages/app/src/i18n/bs.ts
+++ b/packages/app/src/i18n/bs.ts
@@ -947,18 +947,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra wymaga trybu scenariusza wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra wymaga trybu scenariusza intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Tryb scenariusza",
"settings.general.deepagent.prompt.description": "Wybierz, jak kompozytor obsługuje prompt przed wysłaniem.",
"settings.general.deepagent.prompt.direct": "Tryb bezpośredni",
"settings.general.deepagent.prompt.direct.description":
"Wyślij prompt dokładnie tak, jak został napisany, bez przygotowywania szkicu przez DeepAgent.",
- "settings.general.deepagent.prompt.wish": "Tryb wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Tryb intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent najpierw przygotuje pełny prompt, a potem pozwoli go przejrzeć i edytować.",
- "settings.general.deepagent.wishModel.title": "Model trybu wish",
- "settings.general.deepagent.wishModel.description":
- "Wybierz model używany tylko do przygotowywania szkiców promptów w trybie wish.",
+ "settings.general.deepagent.intelligenceModel.title": "Model trybu intelligence",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Wybierz model używany tylko do przygotowywania szkiców promptów w trybie intelligence.",
"settings.general.deepagent.selfLearning.title": "Zatwierdzanie samouczenia",
"settings.general.deepagent.selfLearning.description":
"Zdecyduj, jak wiedza nauczona przez DeepAgent z sesji staje się dostępna do wyszukiwania.",
diff --git a/packages/app/src/i18n/da.ts b/packages/app/src/i18n/da.ts
index 9381ac21..49597235 100644
--- a/packages/app/src/i18n/da.ts
+++ b/packages/app/src/i18n/da.ts
@@ -940,18 +940,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra kræver wish-scenarietilstand.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra kræver intelligence-scenarietilstand.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Scenarietilstand",
"settings.general.deepagent.prompt.description": "Vælg hvordan composeren håndterer din prompt før afsendelse.",
"settings.general.deepagent.prompt.direct": "Direkte tilstand",
"settings.general.deepagent.prompt.direct.description":
"Send din prompt præcis som skrevet, uden at DeepAgent forbereder et udkast.",
- "settings.general.deepagent.prompt.wish": "Wish-tilstand",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Intelligence-tilstand",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent forbereder først en komplet prompt, som du kan gennemse og redigere.",
- "settings.general.deepagent.wishModel.title": "Wish-tilstandsmodel",
- "settings.general.deepagent.wishModel.description":
- "Vælg modellen der kun bruges til at forberede promptudkast i wish-tilstand.",
+ "settings.general.deepagent.intelligenceModel.title": "Intelligence-tilstandsmodel",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Vælg modellen der kun bruges til at forberede promptudkast i intelligence-tilstand.",
"settings.general.deepagent.selfLearning.title": "Godkendelse af selvlæring",
"settings.general.deepagent.selfLearning.description":
"Bestem hvordan viden DeepAgent lærer fra dine sessioner bliver søgbar.",
diff --git a/packages/app/src/i18n/de.ts b/packages/app/src/i18n/de.ts
index b722b2cf..70a5fe6a 100644
--- a/packages/app/src/i18n/de.ts
+++ b/packages/app/src/i18n/de.ts
@@ -881,18 +881,22 @@ export const dict = {
"settings.general.section.deepagent": "DeepAgent",
"settings.general.deepagent.mode.title": "Agent",
"settings.general.deepagent.mode.description": "Wähle die Arbeitsintensität von DeepAgent.",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra erfordert den Wish-Szenariomodus.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra erfordert den Intelligence-Szenariomodus.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Szenariomodus",
"settings.general.deepagent.prompt.description": "Wähle, wie der Composer deinen Prompt vor dem Senden verarbeitet.",
"settings.general.deepagent.prompt.direct": "Direktmodus",
"settings.general.deepagent.prompt.direct.description":
"Sende deinen Prompt genau wie geschrieben, ohne dass DeepAgent einen Entwurf vorbereitet.",
- "settings.general.deepagent.prompt.wish": "Wish-Modus",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Intelligence-Modus",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent bereitet zuerst einen vollständigen Prompt vor, den du prüfen und bearbeiten kannst.",
- "settings.general.deepagent.wishModel.title": "Wish-Modus-Modell",
- "settings.general.deepagent.wishModel.description":
- "Wähle das Modell, das nur für Prompt-Entwürfe im Wish-Modus verwendet wird.",
+ "settings.general.deepagent.intelligenceModel.title": "Intelligence-Modus-Modell",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Wähle das Modell, das nur für Prompt-Entwürfe im Intelligence-Modus verwendet wird.",
"settings.general.deepagent.selfLearning.title": "Selbstlern-Freigabe",
"settings.general.deepagent.selfLearning.description":
"Lege fest, wie von DeepAgent gelernte Sitzungskenntnisse abrufbar werden.",
diff --git a/packages/app/src/i18n/en.ts b/packages/app/src/i18n/en.ts
index c8fa4191..0da53ec3 100644
--- a/packages/app/src/i18n/en.ts
+++ b/packages/app/src/i18n/en.ts
@@ -312,9 +312,9 @@ export const dict = {
"prompt.action.stop": "Stop",
"prompt.scenario.label": "Scenario mode",
"prompt.scenario.direct": "Direct",
- "prompt.scenario.wish": "Wish",
+ "prompt.scenario.intelligence": "Intelligence",
"prompt.scenario.direct.tooltip": "Send your prompt directly",
- "prompt.scenario.wish.tooltip": "DeepAgent prepares a prompt for you",
+ "prompt.scenario.intelligence.tooltip": "DeepAgent prepares a prompt for you",
"prompt.toast.pasteUnsupported.title": "Unsupported attachment",
"prompt.toast.pasteUnsupported.description": "Only images, PDFs, or text files can be attached here.",
@@ -658,6 +658,7 @@ export const dict = {
"notification.session.error.fallbackDescription": "An error occurred",
"home.recentProjects": "Recent projects",
+ "home.newChat": "New chat",
"home.empty.title": "No recent projects",
"home.empty.description": "Get started by opening a local project",
"home.title": "Home",
@@ -716,6 +717,8 @@ export const dict = {
"session.subagents.empty": "No subagents for this session",
"session.subagents.running": "running",
"session.subagents.idle": "idle",
+ "session.fork.derivedFrom": "Derived from {title}",
+ "session.fork.derivedFromUnknown": "Derived from another session",
"browser.title": "Browser",
"browser.address": "Enter a URL",
"browser.back": "Back",
@@ -936,17 +939,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra requires the wish scenario mode.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra requires the intelligence scenario mode.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description":
+ "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Scenario mode",
"settings.general.deepagent.prompt.description": "Choose how the composer handles your prompt before sending.",
"settings.general.deepagent.prompt.direct": "Direct mode",
"settings.general.deepagent.prompt.direct.description":
"Send your prompt exactly as written, without DeepAgent preparing a draft.",
- "settings.general.deepagent.prompt.wish": "Wish mode",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Intelligence mode",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent prepares a complete prompt first, then lets you review and edit it.",
- "settings.general.deepagent.wishModel.title": "Wish mode model",
- "settings.general.deepagent.wishModel.description": "Choose the model used only to prepare wish-mode prompt drafts.",
+ "settings.general.deepagent.intelligenceModel.title": "Intelligence mode model",
+ "settings.general.deepagent.intelligenceModel.description": "Choose the model used only to prepare intelligence-mode prompt drafts.",
"settings.general.deepagent.selfLearning.title": "Self-learning approval",
"settings.general.deepagent.selfLearning.description":
"Decide how knowledge DeepAgent learns from your sessions becomes retrievable.",
diff --git a/packages/app/src/i18n/es.ts b/packages/app/src/i18n/es.ts
index fda9ba46..45453dd4 100644
--- a/packages/app/src/i18n/es.ts
+++ b/packages/app/src/i18n/es.ts
@@ -953,18 +953,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra requiere el modo de escenario wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra requiere el modo de escenario intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Modo de escenario",
"settings.general.deepagent.prompt.description": "Elige cómo el compositor gestiona tu prompt antes de enviarlo.",
"settings.general.deepagent.prompt.direct": "Modo directo",
"settings.general.deepagent.prompt.direct.description":
"Envía tu prompt tal como está escrito, sin que DeepAgent prepare un borrador.",
- "settings.general.deepagent.prompt.wish": "Modo wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Modo intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent prepara primero un prompt completo y luego te permite revisarlo y editarlo.",
- "settings.general.deepagent.wishModel.title": "Modelo del modo wish",
- "settings.general.deepagent.wishModel.description":
- "Elige el modelo usado solo para preparar borradores de prompts del modo wish.",
+ "settings.general.deepagent.intelligenceModel.title": "Modelo del modo intelligence",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Elige el modelo usado solo para preparar borradores de prompts del modo intelligence.",
"settings.general.deepagent.selfLearning.title": "Aprobación de autoaprendizaje",
"settings.general.deepagent.selfLearning.description":
"Decide cómo el conocimiento que DeepAgent aprende de tus sesiones se vuelve recuperable.",
diff --git a/packages/app/src/i18n/fr.ts b/packages/app/src/i18n/fr.ts
index 4b77c2a5..27c1c913 100644
--- a/packages/app/src/i18n/fr.ts
+++ b/packages/app/src/i18n/fr.ts
@@ -880,19 +880,23 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra nécessite le mode scénario wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra nécessite le mode scénario intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Mode scénario",
"settings.general.deepagent.prompt.description":
"Choisissez comment le compositeur traite votre prompt avant l’envoi.",
"settings.general.deepagent.prompt.direct": "Mode direct",
"settings.general.deepagent.prompt.direct.description":
"Envoyez votre prompt exactement tel qu’il est écrit, sans brouillon préparé par DeepAgent.",
- "settings.general.deepagent.prompt.wish": "Mode wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Mode intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent prépare d’abord un prompt complet, puis vous pouvez le relire et le modifier.",
- "settings.general.deepagent.wishModel.title": "Modèle du mode wish",
- "settings.general.deepagent.wishModel.description":
- "Choisissez le modèle utilisé uniquement pour préparer les brouillons de prompts en mode wish.",
+ "settings.general.deepagent.intelligenceModel.title": "Modèle du mode intelligence",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Choisissez le modèle utilisé uniquement pour préparer les brouillons de prompts en mode intelligence.",
"settings.general.deepagent.selfLearning.title": "Approbation de l’auto-apprentissage",
"settings.general.deepagent.selfLearning.description":
"Décidez comment les connaissances apprises par DeepAgent dans vos sessions deviennent récupérables.",
diff --git a/packages/app/src/i18n/ja.ts b/packages/app/src/i18n/ja.ts
index 976358af..c5902ac7 100644
--- a/packages/app/src/i18n/ja.ts
+++ b/packages/app/src/i18n/ja.ts
@@ -866,18 +866,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultraにはウィッシュシナリオモードが必要です。",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultraにはインテリジェンスシナリオモードが必要です。",
+ "settings.general.deepagent.subagentMode.title": "サブエージェント",
+ "settings.general.deepagent.subagentMode.description": "メインエージェントに対するサブエージェントの作業強度を選択します。",
+ "settings.general.deepagent.subagentMode.inherit": "親から継承",
+ "settings.general.deepagent.subagentMode.downgrade": "1段階下げる",
"settings.general.deepagent.prompt.title": "シナリオモード",
"settings.general.deepagent.prompt.description": "送信前にコンポーザーがプロンプトを処理する方法を選択します。",
"settings.general.deepagent.prompt.direct": "直接モード",
"settings.general.deepagent.prompt.direct.description":
"DeepAgentが下書きを準備せず、入力したプロンプトをそのまま送信します。",
- "settings.general.deepagent.prompt.wish": "ウィッシュモード",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "インテリジェンスモード",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgentが完全なプロンプトを先に準備し、確認と編集を可能にします。",
- "settings.general.deepagent.wishModel.title": "ウィッシュモードのモデル",
- "settings.general.deepagent.wishModel.description":
- "ウィッシュモードのプロンプト下書きだけに使うモデルを選択します。",
+ "settings.general.deepagent.intelligenceModel.title": "インテリジェンスモードのモデル",
+ "settings.general.deepagent.intelligenceModel.description":
+ "インテリジェンスモードのプロンプト下書きだけに使うモデルを選択します。",
"settings.general.deepagent.selfLearning.title": "自己学習の承認",
"settings.general.deepagent.selfLearning.description":
"DeepAgentがセッションから学習した知識を検索可能にする方法を決めます。",
diff --git a/packages/app/src/i18n/ko.ts b/packages/app/src/i18n/ko.ts
index 2f1a8dff..9f05a929 100644
--- a/packages/app/src/i18n/ko.ts
+++ b/packages/app/src/i18n/ko.ts
@@ -859,17 +859,21 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra는 wish 시나리오 모드가 필요합니다.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra는 인텔리전스 시나리오 모드가 필요합니다.",
+ "settings.general.deepagent.subagentMode.title": "서브에이전트",
+ "settings.general.deepagent.subagentMode.description": "메인 에이전트 대비 서브에이전트의 작업 강도를 선택합니다.",
+ "settings.general.deepagent.subagentMode.inherit": "상위에서 상속",
+ "settings.general.deepagent.subagentMode.downgrade": "한 단계 낮추기",
"settings.general.deepagent.prompt.title": "시나리오 모드",
"settings.general.deepagent.prompt.description": "전송 전에 작성기가 프롬프트를 처리하는 방식을 선택하세요.",
"settings.general.deepagent.prompt.direct": "직접 모드",
"settings.general.deepagent.prompt.direct.description":
"DeepAgent가 초안을 준비하지 않고 작성한 프롬프트를 그대로 보냅니다.",
- "settings.general.deepagent.prompt.wish": "Wish 모드",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "인텔리전스 모드",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent가 먼저 완전한 프롬프트를 준비한 뒤 검토하고 편집할 수 있게 합니다.",
- "settings.general.deepagent.wishModel.title": "Wish 모드 모델",
- "settings.general.deepagent.wishModel.description": "Wish 모드 프롬프트 초안 준비에만 사용할 모델을 선택하세요.",
+ "settings.general.deepagent.intelligenceModel.title": "인텔리전스 모드 모델",
+ "settings.general.deepagent.intelligenceModel.description": "인텔리전스 모드 프롬프트 초안 준비에만 사용할 모델을 선택하세요.",
"settings.general.deepagent.selfLearning.title": "자가 학습 승인",
"settings.general.deepagent.selfLearning.description":
"DeepAgent가 세션에서 학습한 지식이 검색 가능해지는 방식을 결정합니다.",
diff --git a/packages/app/src/i18n/no.ts b/packages/app/src/i18n/no.ts
index 2680f0e1..fd095d24 100644
--- a/packages/app/src/i18n/no.ts
+++ b/packages/app/src/i18n/no.ts
@@ -947,18 +947,22 @@ export const dict = {
"settings.general.section.deepagent": "DeepAgent",
"settings.general.deepagent.mode.title": "Agent",
"settings.general.deepagent.mode.description": "Velg arbeidsintensiteten til DeepAgent.",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra kræver wish-scenarietilstand.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra kræver intelligence-scenarietilstand.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Scenarietilstand",
"settings.general.deepagent.prompt.description": "Vælg hvordan composeren håndterer din prompt før afsendelse.",
"settings.general.deepagent.prompt.direct": "Direkte tilstand",
"settings.general.deepagent.prompt.direct.description":
"Send din prompt præcis som skrevet, uden at DeepAgent forbereder et udkast.",
- "settings.general.deepagent.prompt.wish": "Wish-tilstand",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Intelligence-tilstand",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent forbereder først en komplet prompt, som du kan gennemse og redigere.",
- "settings.general.deepagent.wishModel.title": "Wish-tilstandsmodel",
- "settings.general.deepagent.wishModel.description":
- "Vælg modellen der kun bruges til at forberede promptudkast i wish-tilstand.",
+ "settings.general.deepagent.intelligenceModel.title": "Intelligence-tilstandsmodel",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Vælg modellen der kun bruges til at forberede promptudkast i intelligence-tilstand.",
"settings.general.deepagent.selfLearning.title": "Godkendelse af selvlæring",
"settings.general.deepagent.selfLearning.description":
"Bestem hvordan viden DeepAgent lærer fra dine sessioner bliver søgbar.",
diff --git a/packages/app/src/i18n/pl.ts b/packages/app/src/i18n/pl.ts
index 813f1115..a7c17d28 100644
--- a/packages/app/src/i18n/pl.ts
+++ b/packages/app/src/i18n/pl.ts
@@ -869,18 +869,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra wymaga trybu scenariusza wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra wymaga trybu scenariusza intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Tryb scenariusza",
"settings.general.deepagent.prompt.description": "Wybierz, jak kompozytor obsługuje prompt przed wysłaniem.",
"settings.general.deepagent.prompt.direct": "Tryb bezpośredni",
"settings.general.deepagent.prompt.direct.description":
"Wyślij prompt dokładnie tak, jak został napisany, bez przygotowywania szkicu przez DeepAgent.",
- "settings.general.deepagent.prompt.wish": "Tryb wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Tryb intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent najpierw przygotuje pełny prompt, a potem pozwoli go przejrzeć i edytować.",
- "settings.general.deepagent.wishModel.title": "Model trybu wish",
- "settings.general.deepagent.wishModel.description":
- "Wybierz model używany tylko do przygotowywania szkiców promptów w trybie wish.",
+ "settings.general.deepagent.intelligenceModel.title": "Model trybu intelligence",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Wybierz model używany tylko do przygotowywania szkiców promptów w trybie intelligence.",
"settings.general.deepagent.selfLearning.title": "Zatwierdzanie samouczenia",
"settings.general.deepagent.selfLearning.description":
"Zdecyduj, jak wiedza nauczona przez DeepAgent z sesji staje się dostępna do wyszukiwania.",
diff --git a/packages/app/src/i18n/ru.ts b/packages/app/src/i18n/ru.ts
index dc7c0731..f226ad73 100644
--- a/packages/app/src/i18n/ru.ts
+++ b/packages/app/src/i18n/ru.ts
@@ -949,18 +949,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra требует режима сценария wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra требует режима сценария intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Режим сценария",
"settings.general.deepagent.prompt.description": "Выберите, как композер обрабатывает ваш промпт перед отправкой.",
"settings.general.deepagent.prompt.direct": "Прямой режим",
"settings.general.deepagent.prompt.direct.description":
"Отправлять промпт как написано, без подготовки черновика DeepAgent.",
- "settings.general.deepagent.prompt.wish": "Режим wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Режим intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent сначала готовит полный промпт, затем позволяет проверить и отредактировать его.",
- "settings.general.deepagent.wishModel.title": "Модель режима wish",
- "settings.general.deepagent.wishModel.description":
- "Выберите модель, используемую только для подготовки черновиков промптов в режиме wish.",
+ "settings.general.deepagent.intelligenceModel.title": "Модель режима intelligence",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Выберите модель, используемую только для подготовки черновиков промптов в режиме intelligence.",
"settings.general.deepagent.selfLearning.title": "Одобрение самообучения",
"settings.general.deepagent.selfLearning.description":
"Определите, как знания, полученные DeepAgent из ваших сессий, становятся доступными для поиска.",
diff --git a/packages/app/src/i18n/th.ts b/packages/app/src/i18n/th.ts
index c071364c..9b480c26 100644
--- a/packages/app/src/i18n/th.ts
+++ b/packages/app/src/i18n/th.ts
@@ -936,16 +936,20 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra ต้องใช้โหมดสถานการณ์ wish",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra ต้องใช้โหมดสถานการณ์ intelligence",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "โหมดสถานการณ์",
"settings.general.deepagent.prompt.description": "เลือกวิธีที่ตัวเขียนจัดการพรอมป์ก่อนส่ง",
"settings.general.deepagent.prompt.direct": "โหมดส่งตรง",
"settings.general.deepagent.prompt.direct.description": "ส่งพรอมป์ตามที่เขียน โดยไม่ให้ DeepAgent เตรียมฉบับร่าง",
- "settings.general.deepagent.prompt.wish": "โหมด wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "โหมด intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent เตรียมพรอมป์ฉบับสมบูรณ์ก่อน แล้วให้คุณตรวจทานและแก้ไข",
- "settings.general.deepagent.wishModel.title": "โมเดลโหมด wish",
- "settings.general.deepagent.wishModel.description": "เลือกโมเดลที่ใช้เฉพาะเพื่อเตรียมฉบับร่างพรอมป์ในโหมด wish",
+ "settings.general.deepagent.intelligenceModel.title": "โมเดลโหมด intelligence",
+ "settings.general.deepagent.intelligenceModel.description": "เลือกโมเดลที่ใช้เฉพาะเพื่อเตรียมฉบับร่างพรอมป์ในโหมด intelligence",
"settings.general.deepagent.selfLearning.title": "การอนุมัติการเรียนรู้ด้วยตนเอง",
"settings.general.deepagent.selfLearning.description":
"กำหนดว่าความรู้ที่ DeepAgent เรียนรู้จากเซสชันของคุณจะค้นคืนได้อย่างไร",
diff --git a/packages/app/src/i18n/tr.ts b/packages/app/src/i18n/tr.ts
index 5ea53068..5494ead0 100644
--- a/packages/app/src/i18n/tr.ts
+++ b/packages/app/src/i18n/tr.ts
@@ -955,19 +955,23 @@ export const dict = {
"settings.general.section.deepagent": "DeepAgent",
"settings.general.deepagent.mode.title": "Ajan",
"settings.general.deepagent.mode.description": "DeepAgent çalışma yoğunluğunu seçin.",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra, wish senaryo modunu gerektirir.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra, intelligence senaryo modunu gerektirir.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Senaryo modu",
"settings.general.deepagent.prompt.description":
"Göndermeden önce oluşturucunun isteminizi nasıl işleyeceğini seçin.",
"settings.general.deepagent.prompt.direct": "Doğrudan mod",
"settings.general.deepagent.prompt.direct.description":
"DeepAgent taslak hazırlamadan isteminizi yazıldığı gibi gönderin.",
- "settings.general.deepagent.prompt.wish": "Wish modu",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Intelligence modu",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent önce eksiksiz bir istem hazırlar, sonra gözden geçirip düzenlemenizi sağlar.",
- "settings.general.deepagent.wishModel.title": "Wish modu modeli",
- "settings.general.deepagent.wishModel.description":
- "Yalnızca wish modu istem taslaklarını hazırlamak için kullanılan modeli seçin.",
+ "settings.general.deepagent.intelligenceModel.title": "Intelligence modu modeli",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Yalnızca intelligence modu istem taslaklarını hazırlamak için kullanılan modeli seçin.",
"settings.general.deepagent.selfLearning.title": "Kendi kendine öğrenme onayı",
"settings.general.deepagent.selfLearning.description":
"DeepAgent’ın oturumlarınızdan öğrendiği bilginin nasıl erişilebilir olacağını belirleyin.",
diff --git a/packages/app/src/i18n/uk.ts b/packages/app/src/i18n/uk.ts
index 58b18f0a..607a0520 100644
--- a/packages/app/src/i18n/uk.ts
+++ b/packages/app/src/i18n/uk.ts
@@ -980,18 +980,22 @@ export const dict = {
"settings.general.deepagent.mode.xhigh": "xhigh",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra потребує режиму сценарію wish.",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra потребує режиму сценарію intelligence.",
+ "settings.general.deepagent.subagentMode.title": "Subagent",
+ "settings.general.deepagent.subagentMode.description": "Choose the work intensity of subagents relative to the main agent.",
+ "settings.general.deepagent.subagentMode.inherit": "Inherit from parent",
+ "settings.general.deepagent.subagentMode.downgrade": "One level lower",
"settings.general.deepagent.prompt.title": "Режим сценарію",
"settings.general.deepagent.prompt.description": "Выберите, как композер обрабатывает ваш промпт перед отправкой.",
"settings.general.deepagent.prompt.direct": "Прямий режим",
"settings.general.deepagent.prompt.direct.description":
"Отправлять промпт как написано, без подготовки черновика DeepAgent.",
- "settings.general.deepagent.prompt.wish": "Режим wish",
- "settings.general.deepagent.prompt.wish.description":
+ "settings.general.deepagent.prompt.intelligence": "Режим intelligence",
+ "settings.general.deepagent.prompt.intelligence.description":
"DeepAgent сначала готовит полный промпт, затем позволяет проверить и отредактировать его.",
- "settings.general.deepagent.wishModel.title": "Модель режиму wish",
- "settings.general.deepagent.wishModel.description":
- "Выберите модель, используемую только для подготовки черновиков промптов в режиме wish.",
+ "settings.general.deepagent.intelligenceModel.title": "Модель режиму intelligence",
+ "settings.general.deepagent.intelligenceModel.description":
+ "Выберите модель, используемую только для подготовки черновиков промптов в режиме intelligence.",
"settings.general.deepagent.selfLearning.title": "Схвалення самонавчання",
"settings.general.deepagent.selfLearning.description":
"Определите, как знания, полученные DeepAgent из ваших сессий, становятся доступными для поиска.",
diff --git a/packages/app/src/i18n/zh.ts b/packages/app/src/i18n/zh.ts
index 46f2e946..4b217456 100644
--- a/packages/app/src/i18n/zh.ts
+++ b/packages/app/src/i18n/zh.ts
@@ -330,9 +330,9 @@ export const dict = {
"prompt.action.stop": "停止",
"prompt.scenario.label": "情景模式",
"prompt.scenario.direct": "直接",
- "prompt.scenario.wish": "许愿",
+ "prompt.scenario.intelligence": "智能",
"prompt.scenario.direct.tooltip": "直接发送您的提示",
- "prompt.scenario.wish.tooltip": "DeepAgent为您准备提示",
+ "prompt.scenario.intelligence.tooltip": "DeepAgent为您准备提示",
"prompt.toast.pasteUnsupported.title": "不支持的附件",
"prompt.toast.pasteUnsupported.description": "此处仅能附加图片、PDF 或文本文件。",
"prompt.toast.modelAgentRequired.title": "请选择智能体和模型",
@@ -580,6 +580,7 @@ export const dict = {
"notification.session.error.fallbackDescription": "发生错误",
"home.recentProjects": "最近项目",
+ "home.newChat": "新对话",
"home.empty.title": "没有最近项目",
"home.empty.description": "通过打开本地项目开始使用",
"home.title": "主页",
@@ -628,6 +629,8 @@ export const dict = {
"session.subagents.empty": "此会话没有子 Agent",
"session.subagents.running": "运行中",
"session.subagents.idle": "空闲",
+ "session.fork.derivedFrom": "从《{title}》派生",
+ "session.fork.derivedFromUnknown": "从其他对话派生",
"browser.title": "浏览器",
"browser.address": "输入网址",
"browser.back": "后退",
@@ -766,15 +769,19 @@ export const dict = {
"settings.general.deepagent.mode.high": "high",
"settings.general.deepagent.mode.max": "max",
"settings.general.deepagent.mode.ultra": "ultra",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra 需要许愿情景模式。",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra 需要智能情景模式。",
+ "settings.general.deepagent.subagentMode.title": "子智能体",
+ "settings.general.deepagent.subagentMode.description": "选择子智能体相对于主智能体的工作强度。",
+ "settings.general.deepagent.subagentMode.inherit": "继承父级",
+ "settings.general.deepagent.subagentMode.downgrade": "降低一级",
"settings.general.deepagent.prompt.title": "情景模式",
"settings.general.deepagent.prompt.description": "选择输入框在发送前如何处理您的提示。",
"settings.general.deepagent.prompt.direct": "直接模式",
"settings.general.deepagent.prompt.direct.description": "直接发送您写下的提示，不让 DeepAgent 先准备草稿。",
- "settings.general.deepagent.prompt.wish": "许愿模式",
- "settings.general.deepagent.prompt.wish.description": "DeepAgent 会先准备完整提示，再由您确认和编辑。",
- "settings.general.deepagent.wishModel.title": "许愿模式模型",
- "settings.general.deepagent.wishModel.description": "选择仅用于准备许愿模式提示草稿的模型。",
+ "settings.general.deepagent.prompt.intelligence": "智能模式",
+ "settings.general.deepagent.prompt.intelligence.description": "DeepAgent 会先准备完整提示，再由您确认和编辑。",
+ "settings.general.deepagent.intelligenceModel.title": "智能模式模型",
+ "settings.general.deepagent.intelligenceModel.description": "选择仅用于准备智能模式提示草稿的模型。",
"settings.general.deepagent.selfLearning.title": "自学习审批",
"settings.general.deepagent.selfLearning.description": "决定 DeepAgent 从你的会话中学到的知识如何变为可检索。",
"settings.general.deepagent.selfLearning.manual": "人工审批",
diff --git a/packages/app/src/i18n/zht.ts b/packages/app/src/i18n/zht.ts
index db01fb37..f8345e57 100644
--- a/packages/app/src/i18n/zht.ts
+++ b/packages/app/src/i18n/zht.ts
@@ -494,6 +494,7 @@ export const dict = {
"notification.session.error.fallbackDescription": "發生錯誤",
"home.recentProjects": "最近專案",
+ "home.newChat": "新對話",
"home.empty.title": "沒有最近專案",
"home.empty.description": "透過開啟本地專案開始使用",
@@ -936,15 +937,19 @@ export const dict = {
"settings.general.section.deepagent": "DeepAgent",
"settings.general.deepagent.mode.title": "智能體",
"settings.general.deepagent.mode.description": "選擇 DeepAgent 的工作強度。",
- "settings.general.deepagent.mode.ultraRequiresWish": "ultra 需要使用許願場景模式。",
+ "settings.general.deepagent.mode.ultraRequiresIntelligence": "ultra 需要使用智能場景模式。",
+ "settings.general.deepagent.subagentMode.title": "子智能體",
+ "settings.general.deepagent.subagentMode.description": "選擇子智能體相對於主智能體的工作強度。",
+ "settings.general.deepagent.subagentMode.inherit": "繼承父級",
+ "settings.general.deepagent.subagentMode.downgrade": "降低一級",
"settings.general.deepagent.prompt.title": "場景模式",
"settings.general.deepagent.prompt.description": "選擇傳送前輸入器如何處理你的提示。",
"settings.general.deepagent.prompt.direct": "直接模式",
"settings.general.deepagent.prompt.direct.description": "依原文傳送提示，不讓 DeepAgent 先準備草稿。",
- "settings.general.deepagent.prompt.wish": "許願模式",
- "settings.general.deepagent.prompt.wish.description": "DeepAgent 先準備完整提示，再讓你審閱和編輯。",
- "settings.general.deepagent.wishModel.title": "許願模式模型",
- "settings.general.deepagent.wishModel.description": "選擇只用於準備許願模式提示草稿的模型。",
+ "settings.general.deepagent.prompt.intelligence": "智能模式",
+ "settings.general.deepagent.prompt.intelligence.description": "DeepAgent 先準備完整提示，再讓你審閱和編輯。",
+ "settings.general.deepagent.intelligenceModel.title": "智能模式模型",
+ "settings.general.deepagent.intelligenceModel.description": "選擇只用於準備智能模式提示草稿的模型。",
"settings.general.deepagent.selfLearning.title": "自學習審批",
"settings.general.deepagent.selfLearning.description": "決定 DeepAgent 從會話學到的知識如何變成可檢索內容。",
"settings.general.deepagent.selfLearning.manual": "手動審批",
diff --git a/packages/app/src/pages/home.tsx b/packages/app/src/pages/home.tsx
index 12572013..e6e44b11 100644
--- a/packages/app/src/pages/home.tsx
+++ b/packages/app/src/pages/home.tsx
@@ -13,6 +13,8 @@ import { ServerConnection, useServer } from "@/context/server"
import { useServerSync } from "@/context/server-sync"
import { useLanguage } from "@/context/language"
import { useGlobal } from "@/context/global"
+import { sandboxDir } from "@/utils/sandbox"
+import { showToast } from "@/utils/toast"
export default function Home() {
const sync = useServerSync()
@@ -45,6 +47,47 @@ export default function Home() {
navigate(`/${base64Encode(directory)}`)
}
+ // Appendix C 形态二 (form 2): folder-less new chat. Allocate a dedicated sandbox
+ // directory under the server's data dir, materialize it on the server (so the
+ // instance boots against a real path — an absent dir yields empty file trees and
+ // PTY 503s), then route through the existing /:dir route. The sandbox — never "/"
+ // — is what enforces the permission boundary for a project-less chat.
+ async function startFolderlessChat() {
+ const s = server.current
+ if (!s) return
+
+ let directory: string
+ try {
+ directory = sandboxDir(sync.data.path.data)
+ } catch {
+ // Path data not loaded yet — the server is still connecting.
+ showToast({
+ variant: "error",
+ title: language.t("common.requestFailed"),
+ description: language.t("common.loading"),
+ })
+ return
+ }
+
+ const serverCtx = global.createServerCtx(s)
+ // Create a dir-scoped client and ensure the sandbox exists. mkdir({ path: "." })
+ // resolves to the directory itself, which the server ensureDir's — bootstrapping
+ // the sandbox root without needing a pre-existing instance.
+ const client = serverCtx.sdk.createClient({ directory, throwOnError: true })
+ try {
+ await client.file.mkdir({ path: "." })
+ } catch (err) {
+ showToast({
+ variant: "error",
+ title: language.t("common.requestFailed"),
+ description: err instanceof Error ? err.message : String(err),
+ })
+ return
+ }
+
+ openProject(s, directory)
+ }
+
function chooseProject() {
const s = server.current
if (!s) return
@@ -89,9 +132,14 @@ export default function Home() {