axonix
Health Warn
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 5 GitHub stars
Code Pass
- Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
- Permissions — No dangerous permissions requested
This tool is a self-evolving AI agent written in Rust. It runs on a continuous loop designed to autonomously read its own source code, write updates, execute tests, and commit changes to a repository.
Security Assessment
Overall Risk: High. The tool's core functionality relies on autonomous and dynamic code execution, which introduces significant inherent risk. The README explicitly confirms it executes shell commands, utilizing SSH (`/ssh <host> <cmd>`) to manage remote devices and running `cargo test` to verify its self-written code. It also makes extensive network requests, including integrating with third-party APIs like Telegram and Bluesky, and streaming live data to a public dashboard. While a light code scan found no hardcoded secrets or explicitly dangerous patterns, granting an autonomous agent unsupervised SSH access and self-modifying capabilities requires extreme caution.
Quality Assessment
The project is under active development, with recent repository pushes and a comprehensive MIT license. However, it suffers from extremely low community visibility, having only 5 GitHub stars. It lacks a broad user base to independently audit the safety of its autonomous actions or verify the stability of its auto-generated code.
Verdict
Use with caution — while the light code scan is clean, the tool's fundamental design as an unsupervised, self-modifying agent with SSH and API access makes it highly risky for untrusted or production environments.
A self-evolving AI agent that gets more useful every day
Axonix
A Rust coding agent that rewrites itself every few hours — and has been doing so for 8 days.
Axonix started as a fork of yoyo-evolve, a minimal coding agent built on yoagent. The original goal was to make an agent that improves itself autonomously. Eight days in, it has 526 tests, runs on a dedicated Intel NUC, builds real tools for its operator, and maintains a live public dashboard.
This is not a demo. It runs on a cron job. Watch it work: axonix.live
What Axonix Actually Does
A cron job wakes it up every 4 hours. Each session:
- Reads its identity, roadmap, goals, metrics, journal, and open community issues
- Picks what matters — a goal it set for itself, a community request, or a bug it found
- Writes code, runs
cargo test, and commits only if tests pass - Updates its goals and journal, posts to Bluesky, sends a Telegram status update
- Pushes to GitHub — every commit is a real change with a real body
The entire history is in the git log. Nothing is staged or scripted.
What It Has Built (Days 1–8)
These are things Axonix built for itself, shipping one session at a time:
| Capability | When | Details |
|---|---|---|
| Modular Rust CLI | Day 1–2 | Split monolith into cli, render, cost, conversation modules |
| Telegram integration | Day 3 | Two-way: /status, /health, /brief commands; alerts on thresholds |
| Bluesky posting | Day 3 | --bluesky-post flag; posts session announcements automatically |
| Live stream server | Day 4–6 | Session output piped to stream.axonix.live in real time |
| Prediction tracking | Day 4 | PredictionStore, /predict REPL command, calibration data across sessions |
| Structured memory | Day 3 | MemoryStore, /memory REPL command, .axonix/memory.json persisted across sessions |
| Morning brief | Day 4–7 | --brief mode; terminal + Telegram formatted summaries of what matters today |
| Public dashboard | Day 3–5 | Goals, metrics, predictions, and charts at axonix.live |
| Sub-agents | Day 6 | code_reviewer and community_responder wired into every session |
| SSH management | Day 3 | /ssh list and /ssh <host> <cmd> REPL commands for multi-device control |
| GitHub Discussions | Day 4 | Auto-posts journal entries via GraphQL API |
| 500+ tests | Day 7 | Targeted test expansion across under-covered modules; 526 and growing |
The Roadmap
| Level | Theme | Status |
|---|---|---|
| 1 ✅ | Survive | Don't break. Build trust in its own code. |
| 2 → | Know Itself | Metrics, self-assessment, goal formation working. |
| 3 → | Be Visible | Dashboard live. Bluesky, Telegram, stream server running. |
| 4 | Be Useful | Build real tools for the person running it. |
| 5 | Be Irreplaceable | Anticipate needs. Become something nothing generic could replace. |
Boss Level: "I couldn't do without this now."
Talk to It
Open a GitHub issue with the agent-input label. Axonix reads open issues at the start of every session. Issues with more 👍 get prioritized.
Axonix responds as @axonix-bot and closes the issue when the work is done.
What you can send:
- Feature requests — tell it what to build
- Bugs — tell it what's broken
- Challenges — give it a hard task and see what it does
It has a journal, a history of past decisions, and opinions about what it's building. The responses are written by the agent, not by a human.
Run It Yourself
Axonix uses a Claude Pro OAuth token — not the Anthropic API. This means it runs against your subscription rather than per-token billing. The token format is sk-ant-oat01-....
Get your token: claude.ai → Settings → API Keys → create an OAuth token
git clone https://github.com/coe0718/axonix
cd axonix
cp .env.example .env
# Add ANTHROPIC_API_KEY (your OAuth token) and GITHUB_TOKEN to .env
Run with Docker (recommended):
docker compose up stream -d # start the live stream server
docker compose run --rm axonix # run an interactive session
Run without Docker:
ANTHROPIC_API_KEY=sk-ant-oat01-... cargo run
Trigger a full evolution session:
ANTHROPIC_API_KEY=sk-ant-oat01-... ./scripts/evolve.sh
See .env.example for all configuration options including Telegram, Bluesky, and SSH host settings.
Reading the History
- JOURNAL.md — session logs written by the agent
- GOALS.md — what it's working on and what it has completed
- METRICS.md — tokens used, tests passed, files changed, per session
- git log — every change it has made to itself
Built On
yoagent — minimal agent loop in Rust.
Inspired by yoyo-evolve — the project that started it all.
License
MIT
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found