Governed AI agents that actually do the work - in the cloud or on your own machine - across business systems, real browser tabs and real Android apps. Visual builder, MCP server, and official SDKs in 9 languages.
Evaluate LLMs side-by-side. Benchmark AI models and coding agents across providers like OpenAI, Google, Anthropic, DeepSeek, and more. Supports custom tasks, structured JSON responses, tool use, and LLM-as-judge validation. Originally created by Petr Malik as MindTrial.
Batteries-included helpers for building LLM-powered apps with the Vercel AI SDK and OpenRouter — automatic cost/usage tracking and pretty stream logging.
Experimental Rust vector-compression sidecars implementing TurboQuant (ICLR 2026), PolarQuant and QJL research primitives, with exact reranking and workload-scoped benchmark receipts.
Unified LLM client library for Go - one API, 36 providers (Anthropic, OpenAI, Google Gemini, AWS Bedrock, Mistral, Groq, DeepSeek, +29 more), zero external dependencies.
Pi Agent extension that bridges MCP servers to the LLM via two dynamic-context tools (CallMcpTool + FetchMcpResource). Filesystem-is-everything: each MCP server is described by meta.json + tools/*.json, loaded into context on session start.