Agents et multi-agents

OpenCodeReview

alibaba/open-code-review

An open-source AI code review CLI from Alibaba that combines deterministic review pipelines with an LLM agent to produce structured, line-level feedback.

Ouvrir le dépôt officiel ↗ Page d'accueil du projet ↗
★ 18KÉtoiles
⑂ 1,2KForks
73Problèmes ouverts
GoLangue
Apache-2.0Licence
Q@project.QualityScoreScore éditorial

Vue d’ensemble

OpenCodeReview reviews Git diffs or complete files. Its deterministic layer selects and groups files, matches rules, and checks comment placement, while an agent retrieves repository context and evaluates code through a configurable LLM. The built-in multi-language ruleset includes checks related to null pointer errors, thread safety, XSS, and SQL injection. The repository is primarily written in Go and distributes a global CLI through npm.

Fonctionnalités clés

  • Hybrid deterministic pipeline and LLM-agent architecture
  • Precise line-level structured review comments
  • Workspace, branch-range, single-commit, and full-file scan modes
  • Repository context through file reading, code search, and inspection of related changes
  • Related-file bundling with isolated sub-agent contexts and concurrent-review support
  • Fine-grained rule matching based on file characteristics
  • Built-in multi-language rules covering NPE, thread safety, XSS, and SQL injection
  • Saved sessions with resume and comment-export commands
  • Delegation mode for reviews performed by an external coding agent
  • MCP support for extending the agent with external tools
  • Integrations documented for coding agents and CI/CD platforms

Prérequis, installation et démarrage rapide

Install the documented npm package globally with: npm install -g @alibaba-group/open-code-review. This makes the ocr command available globally. The documentation also references install scripts, GitHub Release binaries, and source installation, but their commands are not included in the supplied excerpt.

Utilisation

Configure standard mode with `ocr config provider` and `ocr config model`, then run `ocr review` inside a project. Examples include `ocr review --from main --to feature-branch` for a ref range, `ocr review --commit abc123` for one commit, and `ocr scan --path internal/agent` for a directory scan. Sessions can be listed with `ocr session list`, resumed with `--resume <session-id>`, and queried with `ocr session comments <session-id>`. Delegation workflows begin with commands such as `ocr delegate preview` and `ocr delegate rule src/main.go src/handler.go`.

Compatibilité des modèles et cas d’usage

The repository description states compatibility with OpenAI- and Anthropic-compatible endpoints. Users can select a built-in provider or add a custom provider and then choose a model. Specific supported model names are not stated in the repository metadata. Claude Code, Codex, Cursor, and OpenCode are documented as coding-agent integrations rather than model compatibility claims.

Licence et notes sur les risques

Licensed under Apache-2.0. The README attributes copyright 2026 to Alibaba. Consult the repository license file for the complete terms.

Publication et maintenance

No release notes or version history were supplied. The repository record lists the default branch as main and indicates that the project is not archived.

Firecrawl

firecrawl/firecrawl

★ 161,1KTypeScript

LangChain

langchain-ai/langchain

★ 143,6KPython

RAGFlow

infiniflow/ragflow

★ 86,7KGo