Open Source Software — Designed for Agentic Workflows
We build tools and educational resources at the intersection of developer tooling, data provenance, and agentic systems. Our specific domain is Python ↔ Rust interop: reaching for Rust from Python — bindings, real threads beyond the GIL, and type-system rigor — without giving up what makes Python productive.
We practice Agentic Engineering, not Vibe Coding. A Centaur Developer is a human / agent pair working as one contributor: the human owns judgment — architecture, review, and the merge button — while agents carry throughput. The discipline is the point:
- Every change rides the same rails: branch → test-driven development → all checks green → pull request → CI → human merge. No exceptions for agent-authored code.
- Every bug fix ships a regression test that failed before the fix.
- Zero-warnings policy at merge — lint noise is where real issues hide.
- Agents work on a leash: tool authority is capability-governed and deny-by-default (see agent-bridle), not entrusted to prompt hygiene.
- Agents are credited, not hidden: agent contributions are attributed as
model@harness(e.g.nemotron3:33b@newt-agent) so provenance survives in the history.
The canonical rules, templates, and skills live in the agents repo.
Our house pattern for shipping software: a Rust core with Python reach.
The engine is a Rust workspace — correctness, speed, and real threads beyond
the GIL — exposed to Python through PyO3 and built with
maturin, so pip install delivers
prebuilt wheels and Python users never need a Rust toolchain. The Rust
crates stay independently usable from crates.io; the Python package is a
first-class surface, not a wrapper afterthought. One source of truth, two
ecosystems served.
| Repo | Description |
|---|---|
| rust-for-pythonistas | Learn Rust through Python analogies — from ownership to content-addressable data |
| agents | Rules templates and skills for Centaur Developers (human / agent contributors) |
| Repo | Description |
|---|---|
| newt-agent | Small, fast, local-first agentic coder — vi to Hermes-Agent's emacs |
| agent-bridle | The capability leash for agent tools — a capability-governed tool registry for the Gilamonster agent line |
| agent-mesh | Cryptographic peer-to-peer agent coordination — vi-minimal mesh, no broker |
| hermes-thoon | A fork of hermes-agent made for SPEED |
| monitor-agent | A lightweight systems monitoring agent for operations |
Each repo's README carries the longer story — what it is, why it exists, and how it fits the PyO3 style.
"Computer Science has as much to do with Computers as Astronomy does with Telescopes." — Edsger W. Dijkstra
Every project here is a telescope, not the sky. The code, the tools, the frameworks — they exist to serve something beyond themselves. User data is sovereign. Tools are replaceable. Formats should be plain text. Don't build lock-in.
Centaur Developers welcome — human, agent, or both. The contribution rules
live in agents; this
org's CONTRIBUTING.md
and AGENTS.md point there. All repositories in this org enforce safety
gates: secret scanning, push protection, and CI checks that reject leaked
credentials or private infrastructure details.
