Skip to main content

Comodor

The coding agent that stops making the same mistake.

Fix something it wrote, and it never writes it that way again — in this project or the next one. Not a setting you turn on. Not a file you maintain.

PyPI Python CI Licence

English Docs Persian (Farsi) Docs Arabic Docs Turkish Docs Deutsch Docs Spanish Docs Russian Docs French Docs Zhongwen Docs

comodor.ai · Installation · Documentation


Why this one

Every agent forgets. You correct the same thing on Monday and again on Friday, because the correction went into a conversation that ended.

Comodor watches what you do to its output. Change a name it chose, rewrite a function it wrote, tell it no, use the existing helper — it works out the rule behind the correction, tells you the rule it wrote, and follows it from then on.

◈ learned: naming.functions Name functions in snake_case. 6 of 6 definitions
◈ learned: line.length Keep lines within about 79 characters. 95% of lines are under 79
◈ learned: quotes.style Use double quotes for string literals. 176 of 176 literals

Those are real, read out of a working install.

/progress shows whether that is actually working — corrections per turn, falling or not. A number, not a claim.

Whether it is any good, measured

Most coding agents ask you to take their quality on trust. Comodor has a benchmark: thirteen tasks in thirteen real repositories, judged by programs — the suite goes green, the file parses, the old name is gone from every file. Three attempts each, reported as a rate, because a single run dressed up as "it passes" is a made-up number.

Results are committed under bench/results/, dated and named for the model they are about, as JSON and as a table.

The first version of that table was wrong, and the correction is the point. One careful task scored a model 0/3 for not asking a question — and the model had in fact found the ambiguity, resolved it from the codebase's own convention, and said so. The judge was demanding a question Comodor's prompt tells the model not to ask. A judge with a made-up standard is the same bug as a test with made-up keys, pointed the other way: one passes broken code, the other fails correct work, and both look exactly like a real number. The task was replaced with one whose missing piece genuinely is not in the repository.

The whole table, and how to run it against your own model, is in bench/. Every number there came from a real run.

Everything else it does

Asks instead of guessing When a request reads two ways it settles what it can by reading your code, then puts the rest as one short form — before writing anything. →
Never surprises you Reading is silent. Writing shows a diff. Commands ask. Every change is checkpointed and /undo puts it back. →
Drives a real browser One that runs JavaScript, keeps cookies and can log in — not a page fetcher. →
Uses your screen Mouse and keyboard in any application, with a halo showing where it is about to click before it clicks. →
Runs on your phone Telegram, Slack or WhatsApp — the whole interface as buttons, running in the background whether or not a terminal is open. Read-only until you say otherwise. → · → · →
Runs a model locally Pick one, watch it download, use it with the network unplugged. No key, no account. →
Lives in your editor Speaks ACP, so VS Code, JetBrains and Zed drive it from their own panels. →
Follows your procedures Write a skill once; it loads when the work matches. 147 ready to install. →
Works with any model Sixteen hosted providers and three local runtimes, or anything with an OpenAI-compatible URL. →
Costs less 86% of input tokens served from cache, measured — not estimated. →

One dependency. The HTTP client, the SSE reader, the WebSocket that drives Chrome, the PNG encoder for screenshots, the Telegram client — all written here. Installing Comodor pulls in rich and nothing else.

Install

Recommended installation

Linux / MacOS

curl -fsSL get.comodor.ai | sh

Windows (Powershell)

irm get.comodor.ai | iex

The one-liner above finds a Python, builds an isolated environment, puts comodor on your PATH, and fetches a Python if there is none. Verified on a bare debian:bookworm-slim with nothing installed.

Already have a package manager you like?

Install With uv

uv tool install comodor

[!NOTE] can use : uv pip install comodor

Install With pip

pip install comodor

[!NOTE] linux on python3 use : pip3 install comodor

Install With pipx

pipx install comodor

In Docker — nothing to install

The agent, a real Chromium, git and ripgrep, in one container. It starts the browser interface and prints the address to open:

git clone https://github.com/ifekri/Comodor.git && cd Comodor
export ANTHROPIC_API_KEY=...        # or OPENAI_API_KEY, OPENROUTER_API_KEY, ...
docker compose up
  Comodor is at  http://127.0.0.1:8765/?token=...
  Working in     /work

Open the link. The token is new every run, so use the one from this run — docker compose logs comodor prints it again.

The agent works on ./work; point it somewhere else with COMODOR_WORKDIR=/path/to/project docker compose up. Or without cloning anything:

docker run --rm -it -p 127.0.0.1:8765:8765 \
  -e ANTHROPIC_API_KEY -v "$PWD:/work" ifekri/comodor:latest

Published to both Docker Hub (ifekri/comodor) and GitHub (ghcr.io/ifekri/comodor) on every release.

Full details, including how to reach it from another machine safely: Docker.

Then comodor. Python 3.11 or newer. Six questions the first time, never again.

No API key? comodor --demo runs the whole interface offline. Or pick a local model in the setup and pay nothing, ever.

Coming from OpenClaw or Hermes? The first screen offers to bring your keys and skills across. How that works.

Using it

comodor                                    # the interface
comodor run "fix the failing test" --yes   # one task, no interface
comodor web                                # from a browser
comodor telegram start                     # from your phone
comodor acp                                # from your editor
comodor doctor                             # is everything alright?
/help      every command          /undo      restore the last change
/mode      act · plan · chat      /progress  proof it is improving
/cost      tokens and spend       /computer  let it use your screen
Esc        stop it                F3         cycle mode

Documentation

docs/README.md — organised by what you are trying to do.

Getting started Install, choose a model, first task
The interface Panels, keys, and all 29 commands
How it learns Corrections, rules, and the evidence
What it can do The 13 tools, and when it uses each
Safety and permissions What it asks, and what it never does
From your phone The Telegram bot, and who it answers
From Slack Socket Mode — five minutes, and no public address
From WhatsApp The Cloud API — longer to set up; Telegram does the same thing
Models on your machine Downloading one, running it offline
Configuration Every setting, and what wins
Cost Caching, budgets, paying less
Troubleshooting When something is wrong

Development

git clone https://github.com/ifekri/Comodor.git
cd Comodor
uv venv && uv pip install -e ".[dev]"
uv run pytest -q
uv run ruff check .

Both pass on Linux and Windows before anything is pushed. CONTRIBUTING.md

Licence

MIT — LICENSE. Security: SECURITY.md, please not a public issue.

Metadata

Release files for comodor 1.2.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for comodor 1.2.1
File Size Uploaded Provenance Actions
comodor-1.2.1.tar.gz 1.3 MB With provenance Details for comodor-1.2.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for comodor 1.2.1
File Interpreter ABI Platform Actions
comodor-1.2.1-py3-none-any.whl Python 3 none any Details for comodor-1.2.1-py3-none-any.whl

Total release size: 2.4 MB

Release files / comodor-1.2.1.tar.gz

Download URL comodor-1.2.1.tar.gz
Size 1.3 MB
Tags Source
SHA-256 checksum
How to use checksums
5e3f7a32d08dd6a4cbaa4f289cc3c89d6f867a793c4e4d04d00ab24c08fd8ca4
BLAKE2b-256 checksum
How to use checksums
fca986159e276416fa92df550e10c3c5907988f886d3f43f88081c5aa64b2322
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 7, 2026.

Transparency log

Release files / comodor-1.2.1-py3-none-any.whl

Download URL comodor-1.2.1-py3-none-any.whl
Size 1.1 MB
Tags Python 3
SHA-256 checksum
How to use checksums
ea3975915d0985a538a5f43919c0a56f63cf8c9777be50fa5f340617d709adaf
BLAKE2b-256 checksum
How to use checksums
e0d59bd7344109dc84ff1058c00d8a2a4f9920e47c9048bd3f15795469af1d5a
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 7, 2026.

Transparency log

Security for comodor 1.2.1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for comodor-1.2.1.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 7, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

1.2.1 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page