OpenCollab README

README проекта с github.com/RISE-X-Lab/OpenCollab. Одним файлом markdown: README.md.

OpenCollab. Program how coding agents collaborate. Verify that they did. OpenCollab. Program how coding agents collaborate. Verify that they did.

A multi-agent coding framework with programmable collaboration and a controllable runtime.

arXiv 2609.38345 Project page Latest release License: MulanPSL-2.0 Python 3.10 through 3.14 CI
English · Chinese

OpenCollab follows a rigorous and restrained software engineering philosophy. In version 0.9.3, its harness core in the OpenCollab/opencollab directory, written in under 50,000 lines, provides comprehensive research instrumentation and statistics while outperforming Claude Code, Codex CLI and OpenHands across several coding benchmarks. The core implementations of these mature competitors have codebases spanning well over 100,000 lines. OpenCollab also won first place at the OAIC Agentic Software Factory Hackathon, which featured a $24,000 prize pool.

Pass@1 of six harnesses using GPT-5.6-Luna at max reasoning effort. OpenCollab (Duo) scores 68.91% on SWE-bench Pro, 83.15% on Terminal-Bench 2.1, and 65.49% on DeepSWE. Pass@1 of six harnesses using GPT-5.6-Luna at max reasoning effort. OpenCollab (Duo) scores 68.91% on SWE-bench Pro, 83.15% on Terminal-Bench 2.1, and 65.49% on DeepSWE.

Full results

News

Quick start

uv sync --locked
cp configs/.env.example configs/.env   # then set OPENCOLLAB_API_KEY
uv run opencollab --workspace .

configs/.env accepts any OpenAI-compatible or Anthropic endpoint. Never commit real API keys. The command starts the built-in Self-Collaboration team: the Analyst delegates to independent Coder and Tester workspaces and adopts the Coder's changes before testing. See cooperative teams for adoption, conflicts, and recovery. Custom teams choose their own tools. Type /help in the terminal for navigation, saving and exit controls. To define additional roles, create and select an explicit team file as shown below. To run Duo on a Git repository, use this command.

uv run opencollab workflow run duo --workspace /path/to/repository \
  --task 'Fix the public issue described here.' \
  --args '{"allow_unisolated_shell":true}'

Place run options such as --workspace, --model and --budget after workflow run NAME. To list workflows in another workspace, use workflow list --workspace PATH. Options placed before the workflow subcommand are interpreted as belonging to the root interactive mode and will produce a usage error.

Program the collaboration

OpenCollab Team and Workflow modes OpenCollab Team and Workflow modes

Examples

First place at the OAIC Hackathon. OpenCollab powered the first-place solution at the OAIC Agentic Software Factory Hackathon, which offered a $24,000 prize pool. Explore the winning approach in the ARC-Bench r6 example and build on it for your own applications.

From 24,000 lines to a 239-line example. Edict is a multi-agent project of roughly 24,000 lines. Mini Edict's original implementation recreates its core collaboration protocol in 239 lines of team and workflow code, showing how OpenCollab makes sophisticated multi-agent designs compact and easy to extend.

98.3% on a NeurIPS 2026 benchmark. IO2CodeBench was introduced in FromIO, accepted at NeurIPS 2026. Built on OpenCollab, this example reports a resolve rate of 98.3% (113/115), up from its 78.3% (90/115) fixed-example baseline with the same gpt-6-luna model, a 20-percentage-point gain across the benchmark's 115 non-multimodal tasks. See the results.

Verify it happened

In a Team, the agents decide the handoffs, and a declared team does not guarantee that they happen. In our runs, the lead agent often briefed its teammates and then did the work itself. OpenCollab enforces the declaration and records what each agent did:

From these records we measure Adherence, the share of runs in which the declared organization actually happened. Changing one setting of the same team raised it from 47.2% to as much as 97.2%. See how Adherence is measured.

Learn more

Guide What it covers
Package guide CLI, Python API, architecture, and runtime behavior
Configuration guide Providers, models, and team files
Duo guide Running the dual-coder workflow from the CLI and SDK
Weave guide Ordered sessions, executable checks, bounded repair and continuation
Benchmark results Tokens, cost, the paired Duo–Base test, and how to run an evaluation
Adherence What the runtime records, and how we measure Adherence
OpenCollab-Eval README Running agents on software-engineering benchmarks
Documentation index Skills, migration guides, and design records
CONTRIBUTING.md Development checks, testing, and releases

Citation

If you find OpenCollab useful, please give it a ⭐ and cite our work. Our paper is OpenCollab: A Multi-Agent Coding Framework with Programmable Collaboration and Controllable Runtime. OpenCollab builds on Self-Collaboration:

@misc{hsu2026opencollab,
  author        = {Hsu, Chun-Wah and Gong, Kai and Wu, Yu and Chen, Xianhe and
                   Liu, Mengyang and Li, Jie and Li, Hanyu and Liu, Zhixuan and
                   Tang, Naisheng and Chi, Jiaying and Fan, Ziheng and He, Xuning and
                   Yang, Xiaokang and Jiang, Xue and Dong, Yihong},
  title         = {OpenCollab: A Multi-Agent Coding Framework with Programmable Collaboration and Controllable Runtime},
  year          = {2026},
  eprint        = {2609.38345},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2609.38345}
}

@article{dong2023self,
  author        = {Dong, Yihong and Jiang, Xue and Jin, Zhi and Li, Ge},
  title         = {Self-Collaboration Code Generation via ChatGPT},
  journal       = {ACM Transactions on Software Engineering and Methodology},
  volume        = {33},
  number        = {7},
  pages         = {189:1--189:38},
  year          = {2024}
}

License

OpenCollab is licensed under the Mulan Permissive Software License v2 (MulanPSL-2.0).

Star History

OpenCollab star history chart OpenCollab star history chart

OpenCollab (RISE-X-Lab/OpenCollab)

Ищем в базе…

Что это такое?OpenCollab — это фреймворк для программирования с несколькими ИИ-агентами, которые совместно решают задачи по написанию кода. Роли агентов можно программировать, а среду выполнения — контролировать. Он ориентирован на исследования и включает встроенную статистику и инструменты для экспериментов.
Это фреймворк, библиотека или приложение?Это фреймворк: команды агентов описываются в YAML-файлах, а логика передачи задач между ними — в Python-модулях workflow. Запускать его можно из терминала или через Python API.
Как мне это запустить?Создайте configs/.env с эндпоинтом, совместимым с OpenAI, или Anthropic, и ключом, затем выполните команду opencollab. Она запускает встроенную команду Self-Collaboration, где Analyst передаёт работу независимым рабочим пространствам Coder и Tester.
Какой провайдер моделей и какие ключи мне нужны?Нужен API-ключ для любого эндпоинта, совместимого с OpenAI, или Anthropic. Ключ указывается в configs/.env и никогда не должен попадать в git.
Можно ли уже использовать это в продакшене?Пока нет: версия 0.9.3 помечена как предрелиз v1.0.0 PreRelease. Авторы планируют выпустить 1.0.0 только после принятия их статьи, поэтому относитесь к нему как к предрелизу.
Оно правда соблюдает структуру команды, которую я описываю?Да. Сообщение по необъявленной связи отклоняется и записывается в лог, бюджет токенов каждого агента проверяется перед каждым вызовом модели, а --trace записывает по одной JSONL-записи на каждый шаг.
Чем это отличается от Claude Code или Codex CLI?Авторы утверждают, что ядро их харнесса объёмом менее 50 000 строк превосходит Claude Code, Codex CLI и OpenHands на нескольких бенчмарках по программированию, тогда как ядра этих проектов занимают более 100 000 строк. Это собственные заявления авторов.
Проверяли ли это вне лаборатории?Он занял первое место на хакатоне OAIC Agentic Software Factory Hackathon, призовой фонд которого составлял $24 000. Пример, построенный на его основе, показывает долю решённых задач 98,3% (113/115) на IO2CodeBench.
Могу ли я задать собственные роли агентов?Да. Напишите YAML-файл команды, в котором для каждой роли указаны промпт, модель, инструменты и лимит токенов, или напишите Python-модуль workflow, который сам передаёт задачи между агентами.
Какие есть похожие проекты?hive, ChatDev, GVS5H, DeLM, rrsi, Agensh, ruflo, open-multi-agent
Автор
RISE-X-Lab
Размер
21.6 MB
Звёзды
1.5K
Возраст
6 мес
Коммиты · 90 дней