feat: add Atomic Chat as local OpenAI-compatible provider

[yanalialiuk]

Description

Adds a first-class atomic_chat LLM provider for Atomic Chat — a local desktop app that exposes an OpenAI-compatible HTTP API (default http://127.0.0.1:1337/v1).

Users can run Trae Agent against locally loaded models without cloud API keys. The client uses the same Chat Completions flow as other OpenAI-compatible providers (e.g. OpenRouter, Doubao), not the Ollama-native SDK path.

More Information

Changes

Area	Change
trae_agent/utils/llm_clients/atomic_chat_client.py	New AtomicChatClient + AtomicChatProvider with defaults: base URL http://127.0.0.1:1337/v1, API key noop
trae_agent/utils/llm_clients/llm_client.py	Register LLMProvider.ATOMIC_CHAT (atomic_chat) in the provider switch
trae_config.json.example / trae_config.yaml.example	Example provider + model entries
docs/atomic_chat.md	Setup, CLI, JSON/YAML config, troubleshooting
README.md	CLI example for --provider atomic_chat
tests/utils/test_atomic_chat_client_utils.py	Unit tests for defaults, provider wiring, tool-calling heuristic; optional live test behind SKIP_ATOMIC_CHAT_TEST

Design notes

• OpenAI-compatible: Atomic Chat implements /v1/chat/completions, so OpenAICompatibleClient is reused instead of adding a separate protocol stack.
• Tool calling: supports_tool_calling() uses name heuristics (qwen, llama, gemma, etc.); can be overridden via supports_tool_calling: false in YAML model config.
• Env overrides: ATOMIC_CHAT_API_KEY and ATOMIC_CHAT_BASE_URL follow the existing resolve_config_value pattern ({PROVIDER}_API_KEY / {PROVIDER}_BASE_URL).

Impact

• No breaking changes to existing providers.
• New provider id: atomic_chat

Validation

Prerequisites

1. Install Atomic Chat and load a model.
2. Enable the local API server (default port 1337).
3. Python ≥ 3.12 and project deps (uv sync or equivalent).

curl http://127.0.0.1:1337/v1/models
# Note a model id from the response (e.g. gemma-4-E4B-it-IQ4_XS)

Unit tests

cd trae-agent
uv run python -m unittest tests.utils.test_atomic_chat_client_utils.TestAtomicChatClient -v

Expected: default base URL/API key, LLMProvider.ATOMIC_CHAT wiring, tool-calling heuristic pass.

CLI

trae-cli run "Reply with ok" \
  --provider atomic_chat \
  --model <your-model-id> \
  --model-base-url http://127.0.0.1:1337/v1

Config files

• Copy the atomic_chat entries from trae_config.json.example and trae_config.yaml.example.
• See docs/atomic_chat.md for full YAML agent wiring.

[CLAassistant]

All committers have signed the CLA.