Coding agents · LLM observability · design systems · macOS
We build the parts of the toolchain that usually get skipped: the agent that actually opens the pull request, the runtime underneath it, the trace that explains what an AI system did and what it cost, and the security pass that runs before the model and costs nothing.
Everything below is open source. Most of it runs on your keys and your infrastructure.
|
An autonomous coding agent for GitHub. Label an issue and it opens a pull request. Comment Three deterministic scanners run before the model on every pull request — committed credentials, infrastructure, and the source code itself — with no model call and no token cost. A check run publishes the verdict, so a finding can block a merge. Hosted GitHub App, GitHub Action, or CLI. Nine providers.
|
The Python runtime underneath. A small, explicit runtime for production agents. You bring an LLM; it gives you the loop around it — tool calling, retries, streaming, memory, sessions, a rule-based permission layer, prompt caching, and cost tracking. Then the batteries: 40+ built-in tools, 17 SaaS connectors, MCP servers, RAG, skills, hooks, deep multi-agent orchestration and browser automation. Provider-agnostic by design — the same agent code runs on OpenAI, Anthropic, Bedrock, Vertex, Gemini, Groq, Together, Ollama, or 100+ models through LiteLLM. Swap the model in one line; nothing else changes.
pip install shipit-agent |
Observability for LLM applications. One coherent record of what an AI system did: which prompt ran, what it cost, which tenant it belonged to, which tools it called, what it retrieved — and why it chose what it chose.
PII is masked before anything is persisted or leaves the process, because masking at display time is theatre once the raw value is on somebody else's infrastructure. Events fan out to Langfuse for analysis and to your own database as the system of record, with failures isolated per destination.
Python 3.11+ MIT · zero required dependencies · framework-agnostic
pip install shipit-watcher| shipit-ui-design | Senior UI/UX in your terminal — a Claude Code plugin. Eleven skills, nine commands, a nine-category rubric, and a bias toward rich rather than minimal-by-default. Bootstraps design systems, generates polished components, and iterates visually through screenshot critique loops. TypeScript MIT · Live → |
| ShipIt Palette | Pick a color, ship the palette. Fourteen export formats, a live UI playground, and accessibility tools that check contrast where it actually matters. JavaScript MIT · Live → |
| Snappilot | The free, open-source Snagit alternative for macOS. Capture, annotate, record and OCR — native Swift, on-device, private. Nothing leaves your machine. Swift SwiftUI · Site → |
| Project | Version | Stars | Language | Last commit |
|---|---|---|---|---|
| forge |  |
 |
 |
 |
| shipit_agent |  |
 |
 |
 |
| shipit-watchtower |  |
 |
 |
 |
| shipit-ui-design | — |  |
 |
 |
| ShipIt_Palette | — |  |
 |
 |
| snappilot |  |
 |
 |
 |
Bring your own key. Nothing here proxies your traffic through us or marks up a provider's price. You hold the credentials, you see the bill.
Deterministic where it can be. A model is good at judging whether something is reachable and worth worrying about. It is unreliable at checking four thousand lines the same way twice. So the checks that can be deterministic are, they run first, and they cost nothing.
Not crying wolf is the hard part. A scanner people stop reading is worse than no scanner. A trace nobody trusts is worse than no trace. Most of the work in these tools is in what they don't say.
Own your data. Traces land in your database. Usage lands in your SQLite file. Dashboards are account- or token-gated by default, never public.
Documentation · Forge guide · PyPI
Issues and pull requests are welcome on every repository.