Gaunt Sloth

Stateless AI code review

Code review that can't be talked into approving.

Address the issues, update the requirements, or do both.
There is no third option.

Stateless by design

The slim @gaunt-sloth/review package reviews a diff against your guidelines and requirements, returns a verdict, and forgets. There is no session to steer and no chat to wear down — the next run starts from zero, so the only way to change the outcome is to change the code or the requirements.

Stateless also means parallel. Run several reviews of the same PR from different angles — security, architecture, another team's guidelines — the way you would ask another team to look at the code.

Your team owns it

Runs on-premises. Nothing has to leave your infrastructure.
Use your local model via Ollama, or any model exposed through a LangChain.js-compatible package.
Every prompt is a plain markdown file in your repo. No hidden instructions.
No vendor lock-in — bring your own API keys, or none at all with a local model.

Small by design

@gaunt-sloth/review depends only on @gaunt-sloth/core. No commander, no MCP, no A2A — you add the one LangChain.js provider package you actually use.
The non-assistant packages keep production dependencies to a minimum; the assistant itself is not far behind.
Split into packages — core, tools, review, api, assistant — so you extend the parts you need and skip the rest.

The fat package

gaunt-sloth-assistant

For developers' machines there is the full CLI: review your own diff before pushing (git --no-pager diff | gsloth review), review PRs with requirements pulled from GitHub issues or Jira (gsloth pr 42 12), connect to MCP servers, chat, write code, and script quick automations.

Commands Overview

Not just reviews

DIY

The packages work outside CI too. Pukeko Robot Controller uses @gaunt-sloth/api as its agent backend to let an LLM drive a biped robot through an observe–reason–act loop: webcam in, motor commands out, local Ollama or Claude as the brain.

Providers

An API key from any of these, or a local model:

Anthropic
OpenAI (and OpenAI-format providers)
Google AI Studio and Vertex AI
OpenRouter, Groq, DeepSeek, xAI
Local: LM Studio, Ollama, llama.cpp via OpenAI compatibility
Ollama with JS config (some of the models, see #107 )

* Any other provider supported by LangChain.JS should also work with JS config .