Ordalin
Submit a tool

Website-checked tool profile

CodeAF

Open-source terminal coding agent that splits work into checked parallel tasks on the open models you choose.

Sources checked

Visit website ↗
Pricing
CodeAF itself is free and open source under Apache-2.0. You pay the model provider you connect, or route work through a Codex plan you already have; built-in limits pause runaway spending. Official source 1 ↗ · Official source 2 ↗
Available as
One binary for macOS, Linux and Windows, installed with a one-line script; it can run the work on a remote machine over your own SSH. Official source 1 ↗
Primary group
Coding & Development

What it does

CodeAF is a coding agent that splits a request into tasks, runs each on its own branch in its own copy of the repo, and merges only what passes an independent check.

Official source 1 ↗

Features

  • Pareto crewing. Each task gets a worker, a planner and a checker, and each seat can run on a different model picked for cost and quality. Official source 1 ↗
  • Fire-and-steer tasks. /task hands off a job that starts at once and splits into parts running side by side, each checked, while your conversation stays free. Official source 1 ↗
  • senior-dev specialist. A built-in harness for large changes that works alone for an hour or more from one brief to a checked result on a branch. Official source 1 ↗
  • Standing orders. Rules, scheduled reminders and watches written in plain English, scoped to a chat, a project or everywhere. Official source 1 ↗
  • Headless mode for other agents. Claude Code or Codex can call codeaf do with --json and get back one result and an exit code. Official source 1 ↗
  • Rewind. /rewind undoes the conversation, and with furrow the files rewind too, including ones git never tracked. Official source 1 ↗

Pricing & free limits

Open source

FreeApache-2.0

The full agent; model calls are billed by the provider you connect. Official source 1 ↗

Billing details

A $500 daily backstop pauses and asks, any plan estimated over $100 waits for approval, and you can set a tighter limit per conversation. Codex plan calls show no dollar price and stop when the plan's usage limit is reached. Official source 1 ↗ · Official source 2 ↗

Free access & limits

Free access and its limits were not established by the checked official sources.

Best for & limitations

Replacing Claude Code, Codex or OpenCode with an agent that runs several checked tasks in parallel on cheaper open models. Official source 1 ↗

Handing off a cross-file fix or migration to run unattended on a branch until its tests pass. Official source 1 ↗

  • Bring your own model. There is no hosted model; you need a provider key, a local Ollama setup or a Codex plan, and your code goes to whichever provider you pick. Official source 1 ↗
  • Vendor benchmark. The accuracy and cost claims come from AgentField's own DeepSWE run on 113 tasks with one open model. Official source 1 ↗
  • Telemetry on by default. Telemetry and the Model Pool send model-performance counts, not code, unless you switch them off. Official source 1 ↗
CodeAF product screenshot
Website previewFixed desktop viewport · 1440 × 900

Similar tools

Published tools that share its categories.