feat(providers): add TARS (Tetrate Agent Router Service) as a built-in inference provider

[zhaohuabing]

Summary

Adds TARS (Tetrate Agent Router Service) as a built-in inference provider, following the DeepInfra precedent (#1902). TARS is a multi-tenant LLM traffic router that fronts ~200 models across upstream providers behind one endpoint, with server-side model routing, fallback chains, BYOK, budget enforcement, and per-key usage attribution.

One tars provider entry gives sandboxes all of that through inference.local, with the TARS key injected at the supervisor proxy and never present in the sandbox environment.

Related Issue

Changes

• crates/openshell-core/src/inference.rs — TARS_PROTOCOLS, TARS_PROFILE, normalize/profile arms, 3 new tests (+ tars added to the OpenAI-embeddings contract test)
• providers/tars.yaml — Providers v2 profile (inference category, bearer credential, SaaS endpoint for policy injection)
• crates/openshell-providers — discovery spec (TARS_API_KEY, AGENTROUTER_API_KEY), registry + built-in YAML registration
• crates/openshell-core/src/telemetry.rs + crates/openshell-server/src/grpc/provider.rs — Tars telemetry bucket
• crates/openshell-server/src/inference.rs — supported-types error message
• Docs: providers-v2.mdx, manage-providers.mdx, architecture/gateway.md

Testing

• mise run pre-commit passes
• Unit tests added/updated
• E2E tests added/updated (if applicable)
• Tested locally on my dev machine:

~/workspace/OpenShell/target/debug/openshell --gateway-endpoint http://127.0.0.1:17690 provider create --name tars --type tars --credential TARS_API_KEY=$TARS_API_KEY
~/workspace/OpenShell/target/debug/openshell --gateway-endpoint http://127.0.0.1:17690 inference set --provider tars-real --model o3-mini
~/workspace/OpenShell/target/debug/openshell --gateway-endpoint http://127.0.0.1:17690 sandbox create --name tars-test
~/workspace/OpenShell/target/debug/openshell --gateway-endpoint http://127.0.0.1:17690 sandbox exec -n tars-test -- curl -s https://inference.local/v1/chat/completions -H 'Content-Type: application/json' -d '{"model":"anything","messages":[{"role":"user","content":"hi from openshell"}]}'
{
  "id": "chatcmpl-E6sodV8LIYMEaQDwhH6MXZ71IljqA",
  "object": "chat.completion",
  "created": 1785309387,
  "model": "o3-mini-2025-01-31",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hi there! How can I help you today?",
        "refusal": null,
        "annotations": []
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 10,
    "completion_tokens": 151,
    "total_tokens": 161,
    "prompt_tokens_details": {
      "cached_tokens": 0,
      "audio_tokens": 0
    },
    "completion_tokens_details": {
      "reasoning_tokens": 128,
      "audio_tokens": 0,
      "accepted_prediction_tokens": 0,
      "rejected_prediction_tokens": 0
    }
  },
  "service_tier": "default",
  "system_fingerprint": "fp_57f03b1976"
}

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.