Back to directory

loomcycle

A self-hostable agentic runtime that runs as one Go binary alongside your application, handling the agent loop, multi-provider LLM routing, memory, and MCP on both sides, reachable over HTTP, gRPC, MCP, or TypeScript/Python adapters.

loomcycle

loomcycle is a sidecar agentic runtime: one roughly 50 MB Go binary that runs next to an application rather than embedded inside it. The host application calls loomcycle over HTTP, gRPC, MCP, or a TypeScript or Python adapter, while the agent loop, multi-provider model routing, memory and channel primitives, MCP server identity, OpenTelemetry tracing, and multi-replica coordination all live inside the binary, so the application itself can stay in whatever language it was written in.

Providers include Anthropic, OpenAI, DeepSeek, Gemini, and Ollama (cloud and local) over native HTTP behind one interface, with config-driven routing, per-tier fallback cascades, model aliases, and a deterministic zero-token-cost synthetic provider for testing. Built-in tools cover file read/write/edit, search, HTTP, web fetch and search, shell execution, and memory/recall.

The project describes itself as past v1.0, feature-complete, and validated by an 8-hour soak test of 1.27 million circuits and 3.8 million agent runs with zero leaks; it distributes through Homebrew, multi-arch Docker images, a Claude Code plugin, and adapters, under Apache-2.0.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.