← Back to projects

Pacenote

active

An open-source Python framework for building coding agents. Pick your provider, pick your tools, pick your loop — then race whole agents against each other on the same task.

PythonCoding AgentsBenchmarksModalTUIMulti-Agent

Pacenote

An open-source Python framework for building coding agents. Pick your provider, pick your tools, pick your loop, and it wires them together. Recreate SWE-Agent, Aider, Cline or Codex in a few lines each, or design something that doesn’t exist yet.

Shipping as chimera-run on PyPI while the rename lands — the repository and package still carry the old name.

Why

Every coding agent is the same four parts: a provider, a set of tools, a reasoning loop, and an environment where code runs. Monoliths bolt those together and make each one impossible to swap. Decomposing them turns “what if this agent used a different loop?” into a one-line experiment instead of a fork of a 50,000-line repo.

What’s in it

  • Seven sibling CLIs built from the same primitives, each with its own character
  • 27 benchmark adapters and a grid runner — thirteen agents across seven benchmarks
  • A TUI multiplexer that races N complete agents on one task, each in an isolated git worktree, with a live scoreboard
  • Modal integration for fan-out: thousands of graded task attempts as concurrent cloud functions
  • An honest harness — errored runs are not graded as passes

Subscribe to the garden

New notes as they sprout — no spam, unsubscribe anytime.