Summary
- Provides a self-hostable TypeScript gateway for LLM, media, search, local, and cloud-agent providers.
- Exposes a local OpenAI-compatible API endpoint and also supports Claude, Gemini, and Responses-style compatibility.
- Routes requests through automatic or custom model combos using strategies based on health, quota, cost, latency, context, and other factors.
- Uses circuit breakers, connection cooldowns, model lockouts, and provider fallback to improve request resilience.
- Offers token-compression capabilities intended to reduce prompt and tool-output token use.
- Targets AI coding tools and agents including Claude Code, Codex, Cursor, Cline, Copilot, Aider, Goose, OpenCode, Qwen Code, and Gemini CLI.
- Includes MCP and A2A capabilities, optional persistent memory, guardrails, and routing/usage telemetry.
- README-stated provider counts, free-tier budgets, and availability are mutable and can change with provider quotas, regions, and terms.