Bring Your Own Model: One FOSS Daemon for VS Code, MCP, and the OpenAI API

Every editor, CLI, and agent wants its own LLM integration different SDKs, duplicate API keys, and no consistent way to run fully local models next to cloud providers.

Abbenay is an MIT-licensed daemon and library that centralizes provider configuration, secrets, and model discovery on your machine. Use Ollama, LM Studio, or cloud APIs through one stack; clients connect via gRPC, HTTP/SSE, an OpenAI-compatible API, and MCP.

Live demo: start the daemon and web UI; register a local model; pick it in VS Code’s Language Model API; point an OpenAI-compatible tool at the same backend; show MCP tool routing with sensible approval defaults.

Architecture (brief): @abbenay/core as an embeddable library vs the full daemon; why loopback binding and auth-by-default matter for local AI (not “air gap = secure by default”).

Who it’s for: students and contributors who already use AI in the IDE but want one config, offline option, and no vendor lock-in, applicable beyond any single product ecosystem.

Contributing: where to start in the repo (providers, docs, tests, VS Code extension), you don’t need to be an ML researcher to help.

Run one local FOSS daemon instead of re-wiring every app to a different LLM SDK.

Configure bring-your-own-model flows (including local Ollama) and reuse the same setup across VS Code, scripts, and OpenAI-compatible clients.

Understand how MCP fits into a local AI stack without handing keys to every extension.

Know the security basics for local LLM gateways (auth, binding, when HTTP auth can be disabled safely).

Leave with a clear path to try Abbenay today and contribute (docs, providers, integration tests).