Product

Framestack AI Builder

A visual builder for Python that keeps your real code in charge. Lay out your app as a graph, adjust each piece, run it, and ship it as a plain Python project you own outright. The canvas always mirrors your actual code — never a stand-in for it.

Get started Quick start v0.1.0 · MIT · desktop app
flows / rag-agent
Framestack AI Builder canvas showing a RAG pipeline feeding an agent node

Inspector

Agent
model
provider / model
system prompt
tools
mcp: search
python: lookup_order

Node registry

27 node kinds, in 8 families

Every block on the canvas is real, working code you can read, run and keep. Framestack supports what's in its library — not “every framework under the sun.”

FastAPI
Serve the graph over HTTP: a service with routers, routes and settings.
service · routers · routes · settings
LangGraph
A reasoning loop over shared state, with nodes and callable tools.
agent · state · nodes · tools · settings
RAG
A retrieval pipeline broken into stages you can inspect and run.
pipeline · chunking · embedding · retrieval · generation
MCP
Publish your own Model Context Protocol server and tools, or consume a third-party server.
your server · your tools · consumed server
Queues & background work
Move work off the request path with Celery.
Celery workers · background tasks
Containers
Package and orchestrate with Docker and Compose.
Docker · Compose
Database
A relational database the graph can read from and write to.
database
Vector store
A vector store for embeddings and retrieval.
vector store

In the workspace

More than a diagram — a place to work

See how it all connects

Framestack shows how your pieces fit together — and, once you run it, how work actually flows between them.

Edit settings in place

Change a node's settings right on its card or in the side panel, and Framestack updates your code for you.

Talk to what you built

Chat with the assembled agent or RAG pipeline from inside the app, and hand documents to a pipeline for indexing.

Run it, stop it, watch it

Start and stop the app, workers and Compose services. A terminal lives inside the app for everything else.

Blueprint catalog

Drop in ready-made pieces that already come with working code — instantly, at no cost.

Agent that writes to your rules

An integrated AI agent writes code that follows your project's conventions. It needs Claude Code installed.

Requirements

What you need to run it

Runs on your machine

It's a desktop app that runs entirely on your own computer, with a Python core doing the real work. Nothing lives in the cloud.

Build from source

A Rust toolchain is required to build. The .app and .dmg are built on macOS.

For the agent half

The integrated code-writing agent needs Claude Code installed on your machine.

Free and open source

MIT-licensed, version 0.1.0, 578 tests and a single gate that CI runs. No account, no cloud.

Quick start

Clone it and open a canvas

Framestack is a desktop app you build from source. Grab the repository, build it with a Rust toolchain, and point it at a Python project — new or one you already have.

Works against an existing Python project.
No account and no cloud — everything is local.
Run strip any time to leave with plain Python.
terminal
$ git clone https://github.com/ysz7/framestack-ai-builder
$ cd framestack-ai-builder

# requires a Rust toolchain; .app / .dmg build on macOS
# exact build steps live in the README

$ framestack open ./my-python-project

Read the full setup in the README →

Prove your first node.

Open source. Run it locally in a few minutes.