Aelix is an agent runtime and extension platform in pure Python. Self-hosted and auditable, extensible in the language your team already writes, on the model budgets you already pay for. It never phones home.
A tool is a plain function — no plugin language, no out-of-process bridge. Import
pandas, an internal SDK, or a warehouse client and hand results straight back to the
model, in-process. Hot-reload with /reload, no restart.
Browse the
extension catalog →
The official catalog Aelix reads by
default. It ships empty and open for submissions.
Hand-written native adapters for Anthropic, OpenAI, Gemini/Vertex, OpenRouter, Cloudflare, and GitHub Copilot — including the seat you already sign in with. Route cheap work and hard reasoning to different models from one session.
A full Ed25519 provenance toolchain with SHA-256 pinning on extensions
(keygen | sign | trust), enforced fail-closed when you install with
--require-signature — off by default, since no first-party keys are
provisioned yet. Plus an offline catalog and an --offline mode that skips
the first-use rg/fd download and catalog fetches. Policy and
guardrails are auditable built-in extensions.
█████╗ ███████╗██╗ ██╗██╗ ██╗ ██╔══██╗██╔════╝██║ ██║╚██╗██╔╝ ███████║█████╗ ██║ ██║ ╚███╔╝ ██╔══██║██╔══╝ ██║ ██║ ██╔██╗ ██║ ██║███████╗███████╗██║██╔╝ ██╗ ╚═╝ ╚═╝╚══════╝╚══════╝╚═╝╚═╝ ╚═╝ Aelix Agent Runtime small kernel | extension platform | policy-first execution ❯ aelix