Active
A self-hosted creative AI workspace for Ollama and cloud models, with plugins, RAG, artifacts, SSO, and a desktop app.
- Ollama + 10+ providers
- RAG, artifacts + plugins
- Apache 2.0
The Libre ecosystem · 5 projects
Small systems.
Full control.
From a local AI workspace to a language model trained from scratch. Each project asks the same question: how much of the stack can a small team own?
Read the field notesOpen tools
Interfaces and agents you can inspect, change, and run yourself.
Early release
A terminal-native coding-agent harness with durable runs, persistent memory, provider routing, browser tools, and a Telegram bridge.
- Textual TUI
- Durable runs + memory
- Multi-provider routing
Operated products
Focused tools for your communications and documentation.
Early access
A lightweight phone desk on Cloudflare Workers and Twilio, covering calls, messaging, voicemail, IVR, and browser calling.
- Calls + SMS/MMS
- Voicemail + IVR
- ≈$1 number + usage
Available
An embeddable AI assistant for documentation sites, designed to add useful product-aware chat in a few minutes.
- Drop-in widget
- Sources + citations
- Five-minute setup
Model research
Training from scratch, publishing the evidence, and learning from what breaks.
Pretraining complete · SFT pending
A 1B-parameter multilingual language model pretrained from scratch on two RTX 4090s, with a custom tokenizer and data pipeline.
- 1B parameters
- 80,004 steps
- ≈42B tokens processed
- 2 × RTX 4090
Working materials
The tools behind the tools.
- Python
- TypeScript
- React
- Node.js
- Cloudflare Workers
- PyTorch
- Twilio
- Docker
- SQLite