feat: add Atlas Cloud as an OpenAI-compatible LLM provider

[lucaszhu-hue]

Summary

Adds Atlas Cloud as a first-class OpenAI-compatible LLM provider for Hindsight.

Atlas Cloud is a full-modal, OpenAI-compatible AI inference platform that gives one API access to DeepSeek, Qwen, GLM, Kimi, MiniMax and more. Because its /v1/chat/completions endpoint is OpenAI-compatible, it drops straight into the existing OpenAICompatibleLLM path — exactly like the deepseek / zai / opencode-go providers already in the tree.

Enable it for fact extraction, reflection and consolidation with:

export HINDSIGHT_API_LLM_PROVIDER=atlas
export HINDSIGHT_API_LLM_API_KEY=your-atlascloud-api-key   # base_url defaults to https://api.atlascloud.ai/v1
export HINDSIGHT_API_LLM_MODEL=deepseek-ai/deepseek-v4-pro

deepseek-ai/deepseek-v4-pro is a reasoning model, so it needs enough max_tokens (>= 512) — otherwise tokens get spent on the chain of thought and content can come back empty.

Changes

• engine/llm_wrapper.py — register "atlas" in create_llm_provider(), in LLMProvider's valid_providers, and add its default base URL (https://api.atlascloud.ai/v1).
• engine/providers/openai_compatible_llm.py — register "atlas" in valid_providers, default base URL, and the API-key-required check.
• config.py — PROVIDER_DEFAULT_MODELS["atlas"] = "deepseek-ai/deepseek-v4-pro".
• hindsight-embed control center — add Atlas Cloud to the config-wizard provider catalog.
• docs — add Atlas Cloud to llmProviders.json (drives the providers grid + table) and a config example in developer/models.mdx.
• README + .env.example — document the new provider.

No new dependencies; reuses the existing OpenAI-compatible client path.

Testing

Verified end-to-end through Hindsight's own code path: instantiated the provider via create_llm_provider(provider="atlas", ...) (confirmed it returns OpenAICompatibleLLM, resolves base_url=https://api.atlascloud.ai/v1 and the default model) and made a live call() to deepseek-ai/deepseek-v4-pro — HTTP 200, correct content, valid token usage. py_compile and JSON validation pass on all touched files.

Note

I help maintain Atlas Cloud and we're happy to support Hindsight users on the platform — glad to adjust naming/defaults or add follow-ups (e.g. embeddings) to match your conventions. Thanks for building Hindsight!

[nicoloboschi]

Provider wiring is clean and mirrors the opencode-go pattern correctly across config.py, llm_wrapper.py, and openai_compatible_llm.py. Approving to merge.