Code review agent powered by DSPy

These details have not been verified by PyPI

Project links

Project description

CodeSpy logo

CodeSpy

An open-source AI reviewer that catches bugs, improves code quality, and integrates directly into your PR workflow, without sacrificing control or security.

Table of Contents
Why CodeSpy?
Features
Installation
Quick Start
Usage
Configuration
Output
- Markdown (default)
- GitHub/GitLab Review Comments
Architecture
DSPy Signatures
Supported Languages
Development
Contributors
License

Why CodeSpy?

Most AI code reviewers are:

❌ Black boxes
❌ SaaS-only
❌ Opaque about reasoning
❌ Risky for sensitive codebases

CodeSpy is different:

🔍 Transparent reasoning
🔐 Self-hostable
🧠 Configurable review rules
🔄 Native PR integration
🧩 Extensible architecture
📦 100% open-source

Built for engineering teams that care about correctness, security, and control.

Features

🔒 Security Analysis - Detects common vulnerabilities (injection, auth issues, data exposure, etc.) with CWE references
🐛 Bug Detection - Identifies logic errors, null references, resource leaks, edge cases
📝 Documentation Review - Checks for missing docstrings, outdated comments, incomplete docs
🔍 Intelligent Scope Detection - Automatically identifies code scopes (frontend, backend, infra, microservice in mono repo, etc...)
💰 Cost Tracking - Track LLM calls, tokens, and costs per review
🤖 Model Agnostic - Works with OpenAI, AWS Bedrock, Anthropic, Ollama, and more via LiteLLM
🐳 Docker Ready - Run locally or in the cloud with Docker
GitHub & GitLab - Works with both platforms, auto-detects from URL
🖥️ Local Reviews - Review local git changes without GitHub/GitLab — diff against any branch, ref, or review uncommitted work
🧩 MCP Server - IDE integration via Model Context Protocol — trigger reviews from AI coding assistants like Cline without leaving your editor
🔌 GitHub Action - One-line integration for automatic PR reviews

Installation

Using pip

pip install codespy-ai

Using Homebrew (macOS/Linux)

brew tap khezen/codespy
brew install codespy

Using Docker

# Pull the pre-built image from GitHub Container Registry
docker pull ghcr.io/khezen/codespy:latest

# Or build locally
docker build -t codespy .

Using Poetry (for development)

# Clone the repository
git clone https://github.com/khezen/codespy.git
cd codespy

# Install dependencies
poetry install

# Or install only production dependencies
poetry install --only main

Quick Start

Get up and running in 30 seconds:

# 1. Set your Git token (or let codespy auto-discover from gh/glab CLI)
export GITHUB_TOKEN=ghp_xxxxxxxxxxxxxxxxxxxx  # For GitHub
# OR
export GITLAB_TOKEN=glpat-xxxxxxxxxxxxxxxxxxxx  # For GitLab

# 2. Set your LLM provider (example with Anthropic)
export DEFAULT_MODEL=anthropic/claude-opus-4-6
export ANTHROPIC_API_KEY=sk-ant-xxxxxxxxxxxxxxxxxxxx

# 3. Review a PR or MR!
codespy review https://github.com/owner/repo/pull/123
# OR
codespy review https://gitlab.com/group/project/-/merge_requests/123

codespy auto-discovers credentials from standard locations (~/.aws/credentials, gh auth token, glab auth token, etc.) - see Configuration for details.

Usage

Command Line

# Review GitHub Pull Request
codespy review https://github.com/owner/repo/pull/123

# Review GitLab Merge Request
codespy review https://gitlab.com/group/project/-/merge_requests/123

# GitLab with nested groups
codespy review https://gitlab.com/group/subgroup/project/-/merge_requests/123

# Self-hosted GitLab
codespy review https://gitlab.mycompany.com/team/project/-/merge_requests/123

# Output as JSON
codespy review https://github.com/owner/repo/pull/123 --output json

# Use a specific model
codespy review https://github.com/owner/repo/pull/123 --model anthropic/claude-opus-4-6

# Use a custom config file
codespy review https://github.com/owner/repo/pull/123 --config path/to/config.yaml
codespy review https://github.com/owner/repo/pull/123 -f staging.yaml

# Disable stdout output (useful with --git-comment)
codespy review https://github.com/owner/repo/pull/123 --no-stdout

# Post review as GitHub/GitLab comment
codespy review https://github.com/owner/repo/pull/123 --git-comment

# Combine: only post to Git platform, no stdout
codespy review https://github.com/owner/repo/pull/123 --no-stdout --git-comment

# Show current configuration
codespy config

# Show configuration from a specific file
codespy config --config path/to/config.yaml

# Show version
codespy --version

# Review local git changes (no GitHub/GitLab needed)
codespy review-local                    # Review current dir vs main
codespy review-local /path/to/repo      # Review specific repo
codespy review-local --base develop     # Compare against develop
codespy review-local --base origin/main # Compare against origin/main
codespy review-local --base HEAD~5      # Compare against 5 commits back

# Review uncommitted changes (staged + unstaged)
codespy review-uncommitted              # Review current dir
codespy review-uncommitted /path/to/repo
codespy review-uncommitted --output json

IDE Integration (MCP Server)

CodeSpy can run as an MCP (Model Context Protocol) server for integration with AI coding assistants like Cline, enabling code reviews directly from your editor without leaving your workflow.

# Start the MCP server
codespy serve

# Use a custom config file
codespy serve --config path/to/config.yaml

Configure your IDE (example for Cline in VS Code):

Add to cline_mcp_settings.json:

{
  "mcpServers": {
    "codespy-reviewer": {
      "command": "codespy",
      "args": ["serve"],
      "env": {
        "DEFAULT_MODEL": "anthropic/claude-opus-4-6",
        "ANTHROPIC_API_KEY": "your-key-here"
      }
    }
  }
}

Or for AWS Bedrock:

{
  "mcpServers": {
    "codespy-reviewer": {
      "command": "codespy",
      "args": ["serve"],
      "env": {
        "DEFAULT_MODEL": "bedrock/us.anthropic.claude-opus-4-6-v1",
        "AWS_REGION": "us-east-1",
        "AWS_ACCESS_KEY_ID": "your-access-key",
        "AWS_SECRET_ACCESS_KEY": "your-secret-key"
      }
    }
  }
}

Available MCP Tools:

review_local_changes(repo_path, base_ref) — Review branch changes vs base (e.g., vs main)
review_uncommitted(repo_path) — Review staged + unstaged working tree changes
review_pr(mr_url) — Review a GitHub PR or GitLab MR by URL

Then ask your AI assistant: "Review my local changes" or "Review uncommitted work in /path/to/repo"

Using Docker

# With docker run (using GHCR image)
docker run --rm \
  -e GITHUB_TOKEN=$GITHUB_TOKEN \
  -e DEFAULT_MODEL=anthropic/claude-opus-4-6 \
  -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \
  ghcr.io/khezen/codespy:latest review https://github.com/owner/repo/pull/123

# Or use a specific version
docker run --rm \
  -e GITHUB_TOKEN=$GITHUB_TOKEN \
  -e DEFAULT_MODEL=anthropic/claude-opus-4-6 \
  -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \
  ghcr.io/khezen/codespy:0.2.1 review https://github.com/owner/repo/pull/123

GitHub Action

Add CodeSpy to your repository for automatic PR reviews:

Trigger on /codespy review comment:

# .github/workflows/codespy-review.yml
name: CodeSpy Code Review

on:
  issue_comment:
    types: [created]

jobs:
  review:
    # Only run on PR comments containing '/codespy review'
    if: |
      github.event.issue.pull_request &&
      contains(github.event.comment.body, '/codespy review')
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write

    steps:
      - name: Run CodeSpy Review
        uses: khezen/codespy@v1
        with:
          model: 'anthropic/claude-opus-4-6'
          anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}

Trigger automatically on every PR:

# .github/workflows/codespy-review.yml
name: CodeSpy Code Review

on:
  pull_request:
    types: [opened, synchronize, reopened]

jobs:
  review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write

    steps:
      - name: Run CodeSpy Review
        uses: khezen/codespy@v1
        with:
          model: 'anthropic/claude-opus-4-6'
          anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}

See .github/workflows/codespy-review.yml.example for more examples.

Configuration

codespy supports two configuration methods:

.env file - Simple environment variables for basic setup
codespy.yaml - Full YAML configuration for advanced options (per-module settings)

Priority: cmd options > Environment Variables > YAML Config > Defaults

Setup

# Copy the example file
cp .env.example .env

Git Platform Tokens

codespy automatically detects the platform (GitHub or GitLab) from the URL and discovers tokens from multiple sources.

GitHub Token

Auto-discovered from:

GITHUB_TOKEN or GH_TOKEN environment variables
GitHub CLI (gh auth token)
Git credential helper
~/.netrc file

Or create a token at https://github.com/settings/tokens with repo scope:

GITHUB_TOKEN=ghp_xxxxxxxxxxxxxxxxxxxx

To disable auto-discovery:

GITHUB_AUTO_DISCOVER_TOKEN=false

GitLab Token

Auto-discovered from:

GITLAB_TOKEN or GITLAB_PRIVATE_TOKEN environment variables
GitLab CLI (glab auth token)
Git credential helper
~/.netrc file
python-gitlab config files (~/.python-gitlab.cfg, /etc/python-gitlab.cfg)

Or create a token at https://gitlab.com/-/user_settings/personal_access_tokens with api scope:

GITLAB_TOKEN=glpat-xxxxxxxxxxxxxxxxxxxx

For self-hosted GitLab:

GITLAB_URL=https://gitlab.mycompany.com
GITLAB_TOKEN=glpat-xxxxxxxxxxxxxxxxxxxx

To disable auto-discovery:

GITLAB_AUTO_DISCOVER_TOKEN=false

LLM Provider

codespy auto-discovers credentials for all providers:

Anthropic (auto-discovers from $ANTHROPIC_API_KEY, ~/.config/anthropic/, ~/.anthropic/):

DEFAULT_MODEL=anthropic/claude-opus-4-6
# Optional - set explicitly or let codespy auto-discover:
# ANTHROPIC_API_KEY=sk-ant-xxxxxxxxxxxxxxxxxxxx

AWS Bedrock (auto-discovers from ~/.aws/credentials, AWS CLI, env vars):

DEFAULT_MODEL=bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0
AWS_REGION=us-east-1
# Optional - uses ~/.aws/credentials by default, or set explicitly:
# AWS_ACCESS_KEY_ID=...
# AWS_SECRET_ACCESS_KEY=...

OpenAI (auto-discovers from $OPENAI_API_KEY, ~/.config/openai/, ~/.openai/):

DEFAULT_MODEL=openai/gpt-5
# Optional - set explicitly or let codespy auto-discover:
# OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxx

Google Gemini (auto-discovers from $GEMINI_API_KEY, $GOOGLE_API_KEY, gcloud ADC):

DEFAULT_MODEL=gemini/gemini-2.5-pro
# Optional - set explicitly or let codespy auto-discover:
# GEMINI_API_KEY=xxxxxxxxxxxxxxxxxxxx

Local Ollama:

DEFAULT_MODEL=ollama/llama3

To disable auto-discovery for specific providers:

AUTO_DISCOVER_AWS=false
AUTO_DISCOVER_OPENAI=false
AUTO_DISCOVER_ANTHROPIC=false
AUTO_DISCOVER_GEMINI=false

Advanced Configuration (YAML)

For per-signature settings, use codespy.yaml. See codespy.yaml for all available options including:

LLM provider settings and auto-discovery
Git platform configuration (GitHub/GitLab)
Per-signature model and iteration overrides
Output format and destination settings
Directory exclusions

Override YAML settings via environment variables using _ separator:

# Default settings
export DEFAULT_MODEL=anthropic/claude-opus-4-6
export DEFAULT_MAX_ITERS=20

# Per-signature settings (use signature name, not module name)
export CODE_REVIEW_MODEL=anthropic/claude-sonnet-4-5-20250929

# Output settings
export OUTPUT_STDOUT=false
export OUTPUT_GIT=true

See codespy.yaml for full configuration options.

Recommended Model Strategy

codespy uses a tiered model approach to balance review quality and cost:

Tier	Role	Default	Recommended Model	Used By
🧠 Smart	Core analysis & reasoning	`DEFAULT_MODEL`	`anthropic/claude-opus-4-6`	Code & doc review, supply chain, scope identification
⚡ Mid-tier	Field extraction	Falls back to `DEFAULT_MODEL`	`anthropic/claude-sonnet-4-5-20250929`	TwoStepAdapter field extraction
💰 Cheap	Summarization	Falls back to `DEFAULT_MODEL`	`anthropic/claude-haiku-4-5-20251001`	PR summary generation

By default, all models use DEFAULT_MODEL (anthropic/claude-opus-4-6). This works out of the box — just set your API credentials and go.

To optimize costs, override the mid-tier and cheap models:

# .env or environment variables
DEFAULT_MODEL=anthropic/claude-opus-4-6                    # Smart tier (default)
EXTRACTION_MODEL=anthropic/claude-sonnet-4-5-20250929      # Mid-tier: field extraction
SUMMARIZATION_MODEL=anthropic/claude-haiku-4-5-20251001    # Cheap tier: PR summary

Or in codespy.yaml:

default_model: anthropic/claude-opus-4-6
extraction_model: anthropic/claude-sonnet-4-5-20250929
signatures:
  summarization:
    model: anthropic/claude-haiku-4-5-20251001

Output

Markdown (default)

# Code Review: Add user authentication

**PR:** [owner/repo#123](https://github.com/owner/repo/pull/123)
**Reviewed at:** 2024-01-15 10:30 UTC
**Model:** anthropic/claude-opus-4-6

## Summary

This PR implements user authentication with JWT tokens...

## Statistics

- **Total Issues:** 3
- **Critical:** 1
- **Security:** 1
- **Bugs:** 1
- **Documentation:** 1

## Issues

### 🔴 Critical (1)

#### SQL Injection Vulnerability

**Location:** `src/auth/login.py:45`
**Category:** security

The user input is directly interpolated into the SQL query...

**Code:**
query = f"SELECT * FROM users WHERE username = '{username}'"

**Suggestion:**
Use parameterized queries instead...

**Reference:** [CWE-89](https://cwe.mitre.org/data/definitions/89.html)

GitHub/GitLab Review Comments

CodeSpy can post reviews directly to GitHub PRs or GitLab MRs as native review comments with inline annotations.

Enable via CLI:

# GitHub
codespy review https://github.com/owner/repo/pull/123 --git-comment

# GitLab
codespy review https://gitlab.com/group/project/-/merge_requests/123 --git-comment

# Combine: only post to platform, no stdout
codespy review https://github.com/owner/repo/pull/123 --no-stdout --git-comment

Enable via configuration:

# Environment variable
export OUTPUT_GIT=true

# Or in codespy.yaml
output_git: true

Features:

🎯 Inline Comments - Issues are posted as review comments on the exact lines where they occur
📏 Multi-line Support - Issues spanning multiple lines are annotated with start/end line ranges
🔴🟠🟡🔵 Severity Indicators - Visual emoji markers for Critical, High, Medium, Low severity
📦 Collapsible Sections - Organized review body with expandable details:
- 📋 Summary of changes
- 🎯 Quality Assessment
- 📊 Statistics table
- 💰 Cost breakdown per signature
- 💡 Recommendation
🔗 CWE References - Security issues link directly to MITRE CWE database

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                           codespy CLI                               │
├─────────────────────────────────────────────────────────────────────┤
│  review <pr_url> [--config ...] [--output json|md] [--model ...]    │
└──────────────────────────────┬──────────────────────────────────────┘
                               │
┌──────────────────────────────▼──────────────────────────────────────┐
│                  Git Platform Integration                           │
│  - GitHub: Fetch PR diff, changed files, commit messages            │
│  - GitLab: Fetch MR diff, changed files, commit messages            │
│  - Auto-detects platform from URL                                   │
│  - Clone/access full repository for context                         │
└──────────────────────────────┬──────────────────────────────────────┘
                               │
┌──────────────────────────────▼──────────────────────────────────────┐
│                   DSPy Review Pipeline                              │
│                                                                     │
│  ┌────────────────────────────────────────────────────────────┐     │
│  │                   Scope Identifier                         │     │
│  │  (identifies code scopes: frontend, backend, infra, etc.)  │     │
│  └──────────────────────────┬─────────────────────────────────┘     │
│                             │                                       │
│  ┌──────────────────────────▼─────────────────────────────────┐     │
│  │              Parallel Review Modules                       │     │
│  │  ┌──────────────┐ ┌──────────────┐ ┌──────────┐            │     │
│  │  │ Supply Chain │ │ Code         │ │   Doc    │            │     │
│  │  │   Auditor    │ │ Reviewer     │ │ Reviewer │            │     │
│  │  │              │ │ (bug+sec+    │ │          │            │     │
│  │  │              │ │  smell)      │ │          │            │     │
│  │  └──────────────┘ └──────────────┘ └──────────┘            │     │
│  └──────────────────────────┬─────────────────────────────────┘     │
│                             │                                       │
│  ┌──────────────────────────▼─────────────────────────────────┐     │
│  │                   PR Summarizer                            │     │
│  │  (generates summary, quality assessment, recommendation)   │     │
│  └────────────────────────────────────────────────────────────┘     │
│                                                                     │
│                     Cost Tracker (tokens, calls, $)                 │
└──────────────────────────────┬──────────────────────────────────────┘
                               │
┌──────────────────────────────▼──────────────────────────────────────┐
│                        Tools Layer                                  │
│  ┌────────────┐  ┌────────────┐  ┌────────────┐  ┌──────────────┐   │
│  │ Filesystem │  │    Git     │  │    Web     │  │  Cyber/OSV   │   │
│  │            │  │ (GH + GL)  │  │            │  │              │   │
│  └────────────┘  └────────────┘  └────────────┘  └──────────────┘   │
│  ┌────────────────────────────────────────────────────────────────┐ │
│  │                      Parsers                                   │ │
│  │  ┌─────────────────┐  ┌────────────────────────────────────┐   │ │
│  │  │     Ripgrep     │  │           Tree-sitter              │   │ │
│  │  │  (code search)  │  │  (multi-language AST parsing)      │   │ │
│  │  └─────────────────┘  └────────────────────────────────────┘   │ │
│  └────────────────────────────────────────────────────────────────┘ │
└──────────────────────────────┬──────────────────────────────────────┘
                               │
┌──────────────────────────────▼──────────────────────────────────────┐
│                      LLM Backend (LiteLLM)                          │
│    Bedrock | OpenAI | Anthropic | Ollama | Any OpenAI-compatible    │
└─────────────────────────────────────────────────────────────────────┘

DSPy Signatures

The review is powered by DSPy signatures that structure the LLM's analysis:

Signature	Config Key	Description
ScopeIdentifierSignature	`scope`	Identifies code scopes (frontend, backend, infra, microservice in mono repo, etc...)
CodeReviewSignature	`code_review`	Detects verified bugs, security vulnerabilities, removed defensive code, and code smells
DocReviewSignature	`doc`	Detects stale or wrong documentation caused by code changes
SupplyChainSecuritySignature	`supply_chain`	Analyzes artifacts (Dockerfiles) and dependencies for supply chain security
MRSummarySignature	`summarization`	Generates summary, quality assessment, and recommendation

Supported Languages

Tree-sitter based parsing for context-aware analysis:

Language	Extensions	Features
Python	`.py`	Functions, classes, imports
JavaScript	`.js`, `.jsx`	Functions, classes, imports
TypeScript	`.ts`, `.tsx`	Functions, classes, interfaces
Go	`.go`	Functions, structs, interfaces
Java	`.java`	Methods, classes, packages
Kotlin	`.kt`	Functions, classes, objects
Swift	`.swift`	Functions, classes, structs
Objective-C	`.m`, `.h`	Methods, interfaces, protocols
Rust	`.rs`	Functions, structs, traits, impl blocks
Terraform	`.tf`	Resources, data sources, modules, variables

All languages are supported for security, bug, and documentation analysis.

Development

# Quick setup (creates .env and installs dependencies)
make setup

# Or manually with Poetry:
poetry install           # Install all dependencies including dev
poetry lock              # Update lock file

# Available make targets
make help

# Run commands with Poetry
make lint                # Run ruff linter
make format              # Format code with ruff
make typecheck           # Run mypy type checker
make test                # Run pytest tests
make build               # Build package with Poetry
make clean               # Clean build artifacts

# Or run directly:
poetry run codespy review https://github.com/owner/repo/pull/123
poetry run ruff check src/
poetry run mypy src/

Contributors

@khezen
@pranavsriram8

License

MIT

Project details

These details have not been verified by PyPI

Project links

Release history Release notifications | RSS feed

This version

0.4.1

Mar 9, 2026

0.4.0

Mar 2, 2026

0.3.2

Feb 21, 2026

0.3.1

Feb 15, 2026

0.3.0

Feb 14, 2026

0.2.3

Feb 9, 2026

0.2.2

Feb 8, 2026

0.2.1

Feb 8, 2026

0.2.0

Feb 7, 2026

0.1.1

Feb 5, 2026

0.1.0

Feb 5, 2026

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

codespy_ai-0.4.1.tar.gz (99.2 kB view details)

Uploaded Mar 9, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

codespy_ai-0.4.1-py3-none-any.whl (132.6 kB view details)

Uploaded Mar 9, 2026 Python 3

File details

Details for the file codespy_ai-0.4.1.tar.gz.

File metadata

Download URL: codespy_ai-0.4.1.tar.gz
Upload date: Mar 9, 2026
Size: 99.2 kB
Tags: Source
Uploaded using Trusted Publishing? No
Uploaded via: twine/6.1.0 CPython/3.13.7

File hashes

Hashes for codespy_ai-0.4.1.tar.gz
Algorithm	Hash digest
SHA256	`3de78db0959094fad78d2ebb58cbe73b8940f82e8f8bd253b5b6a2ca4118b96b`
MD5	`0f6136a6394d18b7b14e61d690d7bd92`
BLAKE2b-256	`b88ce5e8ca6d451501d2efeb1f39eef70740da44d3b1e3a67a23c87b88a84491`

See more details on using hashes here.

File details

Details for the file codespy_ai-0.4.1-py3-none-any.whl.

File metadata

Download URL: codespy_ai-0.4.1-py3-none-any.whl
Upload date: Mar 9, 2026
Size: 132.6 kB
Tags: Python 3
Uploaded using Trusted Publishing? No
Uploaded via: twine/6.1.0 CPython/3.13.7

File hashes

Hashes for codespy_ai-0.4.1-py3-none-any.whl
Algorithm	Hash digest
SHA256	`e041c28ef2279b2162d6406d675a58b59cad70d74e81c6ff2b4f304f83cc629a`
MD5	`47ac94358fa80eb1bf8be13321053f5d`
BLAKE2b-256	`a867dd215847ee617f2ebec67a27da57928634042e19f04689ca60fac99f907c`

See more details on using hashes here.

codespy-ai 0.4.1

Navigation

Verified details

Maintainers

Unverified details

Project links

Meta

Classifiers

Project description

CodeSpy

Table of Contents

Why CodeSpy?

Features

Installation

Using pip

Using Homebrew (macOS/Linux)

Using Docker

Using Poetry (for development)

Quick Start

Usage

Command Line

IDE Integration (MCP Server)

Using Docker

GitHub Action

Configuration

Setup

Git Platform Tokens

GitHub Token

GitLab Token

LLM Provider

Advanced Configuration (YAML)

Recommended Model Strategy

Output

Markdown (default)

GitHub/GitLab Review Comments

Architecture

DSPy Signatures

Supported Languages

Development

Contributors

License

Project details

Verified details

Maintainers

Unverified details

Project links

Meta

Classifiers

Release history Release notifications | RSS feed

Download files

Source Distribution

Built Distribution

File details

File metadata

File hashes

File details

File metadata

File hashes