Important
codeguard has moved into crewforge. It now lives on as crewforge review, on a shared core
that fails loudly instead of reporting success on empty output. This repository is archived and kept for history;
new issues and changes go to crewforge.
Multi-agent code review for any OpenAI-compatible LLM
Six specialist agents review your code in parallel (security, style, complexity, logic, testing, documentation); a seventh merges their output into one prioritized report with an APPROVE / REQUEST_CHANGES / COMMENT recommendation.
Works with OpenAI, OpenRouter, Ollama, llama.cpp, Groq, DeepSeek, Together, Mistral, Xiaomi MiMo, or any endpoint that speaks the OpenAI /chat/completions protocol. Pick a provider with one config line, no code changes. The default is OpenAI.
┌────────────────────────────────────────────────────────────────┐
│ CodeGuard Pipeline │
│ Any OpenAI-compatible LLM backend │
├────────────────────────────────────────────────────────────────┤
│ │
│ Phase 1: PARALLEL REVIEW (6 agents) │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Security │ │ Style │ │Complexity│ │
│ │ Agent │ │ Agent │ │ Agent │ │
│ └──────────┘ └──────────┘ └──────────┘ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Logic │ │ Testing │ │ Docs │ │
│ │ Agent │ │ Agent │ │ Agent │ │
│ └──────────┘ └──────────┘ └──────────┘ │
│ │ │ │ │
│ └──────────┼───────────┘ │
│ ▼ │
│ Phase 2: AGGREGATE │
│ ┌─────────────────────────────────┐ │
│ │ Summary Agent │ │
│ │ Deduplicate · Prioritize · │ │
│ │ Score · Decision │ │
│ └─────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────┐ │
│ │ Token Tracker & Report │ │
│ └─────────────────────────────────┘ │
│ │
│ Protocol: OpenAI-compatible /chat/completions │
│ Auth: bearer token or api-key header (per provider) │
│ Phase 1 runs 6 agents concurrently (asyncio.gather) │
└────────────────────────────────────────────────────────────────┘
| # | Agent | Reviews | Output |
|---|---|---|---|
| 1 | Security | Injection, secrets, crypto, validation | CWE-tagged findings |
| 2 | Style | Naming, formatting, idioms, dead code | Before/after suggestions |
| 3 | Complexity | Cyclomatic/cognitive complexity, duplication | Refactoring suggestions |
| 4 | Logic | Edge cases, null handling, race conditions | Bug scenarios + fixes |
| 5 | Testing | Coverage gaps, missing test cases | Pytest snippets |
| 6 | Documentation | Docstrings, type hints, TODO comments | Suggested docs |
| 7 | Summary | Aggregate, deduplicate, prioritize | APPROVE/CHANGES/COMMENT |
Each agent asks the model for JSON findings; CodeGuard passes the model's answer through as-is, so the exact shape depends on the model you use.
git clone https://github.com/aimanmalib/codeguard.git
cd codeguard
python -m venv .venv && source .venv/bin/activate
pip install -e .
# OpenAI is the default provider: just set the key
export OPENAI_API_KEY="sk-..."
# Review a file (Markdown report in the terminal)
codeguard review --path ./src/main.py
# Review a directory, JSON report to a file
codeguard review --path ./src --format json --output report.json
# List the agents
codeguard agentsA review makes 7 LLM calls (6 in parallel, then the summary). If every review agent fails (wrong key, unreachable endpoint), the command exits with status 1 and prints the first error.
--format is text (rendered Markdown, the default), markdown (raw Markdown on stdout), or json. Status messages go to stderr, so --format json on stdout is pipe-friendly.
Built-in presets (from src/codeguard/core/config.py):
| Provider | provider: |
Default model | Auth header | API key env var |
|---|---|---|---|---|
| OpenAI (default) | openai |
gpt-4o-mini |
Bearer | OPENAI_API_KEY |
| OpenRouter | openrouter |
openai/gpt-4o-mini |
Bearer | OPENROUTER_API_KEY |
| Ollama (local) | ollama |
llama3.1 |
Bearer | OLLAMA_API_KEY (optional) |
| Groq | groq |
llama-3.3-70b-versatile |
Bearer | GROQ_API_KEY |
| DeepSeek | deepseek |
deepseek-chat |
Bearer | DEEPSEEK_API_KEY |
| Together | together |
meta-llama/Llama-3.3-70B-Instruct-Turbo |
Bearer | TOGETHER_API_KEY |
| Mistral | mistral |
mistral-small-latest |
Bearer | MISTRAL_API_KEY |
| Xiaomi MiMo | mimo |
mimo-v2.5-pro |
api-key |
MIMO_API_KEY |
Every preset's base URL can be overridden with <PROVIDER>_BASE_URL (for example OPENAI_BASE_URL, OLLAMA_BASE_URL) or with base_url in the config file. That is also how you use any other OpenAI-compatible server (llama.cpp, vLLM, LM Studio, a local proxy).
Without --config, CodeGuard uses the openai preset and reads OPENAI_API_KEY / OPENAI_BASE_URL from the environment. To pick another provider or model, pass a YAML file:
# codeguard.yaml
llm:
provider: openrouter # any preset name from the table above
model: openai/gpt-4o # optional; defaults to the preset's model
# base_url: http://localhost:8080/v1 # optional; any OpenAI-compatible endpoint
temperature: 0.3 # low for consistent severity classification
max_tokens: 4096
max_retries: 3 # retries on 429 / 5xx / network errorsexport OPENROUTER_API_KEY="sk-or-..."
codeguard --config codeguard.yaml review --path ./srcThe API key is read from the provider's env var. You can also set api_key: in the file, but keep such a file out of version control. Environment variables are not expanded inside the YAML.
pip install -e ".[dev]"
ruff check src/ tests/ && ruff format --check src/ tests/
pytestThe test suite mocks all HTTP calls, so it needs no API key. CI runs it on Python 3.10 to 3.13.
src/codeguard/
├── cli.py # `codeguard` command (review, agents)
├── agents/ # 6 review agents + summary agent, registry in __init__.py
├── core/
│ ├── config.py # provider presets, LLMConfig, YAML loading
│ ├── llm_client.py # async OpenAI-compatible client (retries, usage tracking)
│ ├── mimo_client.py # backward-compat import shim for llm_client
│ └── token_tracker.py # per-agent token and latency metrics
└── pipeline/
└── orchestrator.py # parallel review phase, then summary
tests/unit/ # unit tests (HTTP mocked)
See CONTRIBUTING.md. Please report vulnerabilities privately, as described in SECURITY.md.
MIT © 2026 Muhammad Aiman
Built by Sasa Teguh · Malaysia