cardchase/loomflow
cardchase/loomflow has 55 GitHub stars and 5 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-15 18:03 UTC. Last push 2026-09-25.
Measured star growth
- in 24h
- 0
- in 7 days
- Update delayed
- in 15 days
- Update delayed
- in 1 month
- Update delayed
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
A self-hosted, lightweight, visual ETL (Extract, Transform, Load) platform inspired by Alteryx.
DataREADME.md
Original⛵ Loomflow (Beta: Local AI Branch)
A self-hosted, visual ETL and workflow synthesizer built on Polars, React Flow, and Local LLMs.
[!WARNING] 🚧 ACTIVE DEVELOPMENT (
betabranch) You are currently viewing thebetabranch. This branch is dedicated to our transition to a 100% local, private, and self-hostable AI architecture.We are decoupling all agentic synthesis features from proprietary cloud APIs. Our primary objective here is building native support for local open-weights models (via Ollama, LM Studio, llama.cpp) using standardized OpenAI-compatible local endpoints. Your data and schemas should never leave your machine.
🎯 The Vision: Autonomous, Local-First Data Engineering
Loomflow brings the drag-and-drop visual pipeline building of massive enterprise tools directly to your local machine, powered by the sub-millisecond execution of Polars.
On this branch, we are building the Autonomous Workflow Synthesizer. Instead of manually dragging nodes, you give a local LLM a natural language objective. The agent uses an In-Memory Sandbox to profile data slices, test transformations, catch its own exceptions, and then automatically compile the verified execution trace into a React Flow DAG on your canvas.
Why Local?
If you are dealing with company logs, CRM support tickets, or lab data, sending your schema and data to cloud LLMs is a non-starter. Loomflow's synthesizer is being built to run entirely on local consumer hardware (Llama 3, Qwen, Gemma) so you retain complete data sovereignty.
🛠️ Beta Roadmap & How to Contribute
We are actively looking for contributors to help us build out the multi-domain sandbox and local model orchestration. Pick a task below and open a PR!
1. Provider-Agnostic LLM Orchestrator (backend/agent/)
- Implement the ReAct loop supporting local OpenAI-compatible endpoints (
http://localhost:1234/v1). - Map Python Pydantic tool definitions to standard JSON schema for local model function calling.
- Build the self-healing error loop (catching Polars exceptions and feeding the stack trace back to the LLM).
2. The Sandbox Engine (backend/agent/sandbox/)
-
SandboxSession: In-memorypl.LazyFramemicro-cache (max 200 rows) for zero-latency agent iterations. -
inspect_source_metadata(): Tool for the LLM to read columns, dtypes, and null ratios. -
test_polars_transformation(): Safe execution environment for the LLM to test its generated nodes before committing them to the canvas.
3. Multi-Domain Canvas Nodes
We are expanding Loomflow beyond basic ETL. Help us build the execution logic for:
- Business & CRM:
SLATimerNode,RuleRouterNode(multi-port branching),LocalTextClassificationNode. - STEM & Math:
SymbolicFormulaNode(integratingsympyfor calculus and equation solving with KaTeX rendering). - Predictive ML:
WindowFeaturesNode,MLPredictorNode(XGBoost/Poisson).
(Developers do NOT need to write React to build nodes! Build a Python class in backend/app/tools/, define a MANIFEST dictionary, and the platform auto-generates the UI).
🚀 Quick Start (Development)
Run both the backend and frontend simultaneously with our automated scripts:
Windows (PowerShell)
.\run.ps1
macOS / Linux (Bash)
Bash
chmod +x run.sh
./run.sh
For manual startup, run uvicorn in the /backend and npm run dev in the /frontend.
🧩 Architecture
Code snippet
graph TD
A[React Canvas UI] -->|Prompt + Data Source| B[Local LLM / LM Studio]
B -->|Tool Call| C[Agent Sandbox Engine]
C -->|Test on 100-row slice| D[Polars In-Memory Eval]
D -->|Error Trace| B
D -->|Verified Trace| E[DAG Layout Compiler]
E -->|JSON Nodes & Edges| A