Skip to content

Daimon

The spirit that runs alongside your AI app.

Daimon is a local sidecar process that gives your application a single, stable HTTP interface to any LLM. Swap providers, rotate keys, add tracing, or wire up MCP tool servers — without touching your application code.

your app  ──POST /v1/converse/claude──▶  daimon  ──▶  Anthropic API
          ◀── text/event-stream ────────────────────────────────────
                                              ▼
                                       MCP tool server(s)

Why Daimon?

Without daimon With daimon
Provider SDKs in every service One HTTP call from anywhere
API keys scattered across code Keys live in one config file
Changing providers = code changes Change a line in YAML
No tracing without instrumentation OpenTelemetry built in
Tool calls require orchestration code MCP agentic loop handled for you

Key features

  • Streaming-first — responses arrive as Server-Sent Events, token by token
  • Provider-agnostic — OpenAI, Anthropic, and any OpenAI-compatible server (Ollama, LM Studio, llama.cpp)
  • Inference parameter defaults — set temperature, max_tokens, system prompt, and more per-component in YAML; override per-request at runtime
  • Server-side sessions — pass a session_id and daimon maintains conversation history for you; clients only send the new turn
  • MCP tool calls — configure MCP servers in YAML; daimon injects their tools into every request and drives the full agentic loop transparently
  • Python SDK — pip install daimon-client for sync and async streaming clients
  • TypeScript SDK — npm install daimon-client for Node.js and edge runtimes
  • OpenTelemetry tracing — structured traces per request, compatible with any OTLP collector

Install

brew tap sonicboom15/tap
brew install daimon

winget install sonicboom15.daimon

scoop bucket add sonicboom15 https://github.com/sonicboom15/scoop-bucket
scoop install daimon

Download from the latest release.

sudo dpkg -i daimon_*_linux_amd64.deb   # Debian / Ubuntu
sudo rpm -i  daimon_*_linux_amd64.rpm   # RHEL / Fedora

First request in 60 seconds

Explore the docs

  • Quick Start

    Up and running in five minutes with a config file and your first request.

    Quick Start

  • HTTP API

    Full reference for /v1/converse, sessions, SSE chunk types, and healthz.

    API reference

  • Python SDK

    pip install daimon-client — sync and async streaming clients with sessions.

    Python SDK

  • TypeScript SDK

    npm install daimon-client — native fetch, async generators, full type safety.

    TypeScript SDK

  • Configuration

    Components, inference defaults, MCP servers, and telemetry — all in one YAML file.

    Configuration

  • Tool Calls (MCP)

    Wire up filesystem, GitHub, search, and custom MCP servers with zero client code.

    MCP tools