Today's generative AI is a sealed machine built on the mathematics of encoded meaning. velella lays that math out on an open canvas: every operator, every wire, every result is there to inspect. Follow any finding back to the data it came from. Trust what you see, because you can see all of it. Find hidden structures in the things you know best and love.
Yours to keep, always
Software should not hold your work hostage. velella is open, and everything you build in it is stored as plain, readable JSON. Whatever you make here, you can always carry out with you.
Your data stays put
velella runs in your browser: no accounts, no cloud storage. Choose a European provider like Mistral and you can keep your embeddings and generation on European, GDPR-friendly ground.
Build your own on top
velella is an engine as much as an app: a stable, open backdrop to build upon. Write your own node pack, bolt on your own operators, run a private instance for your team. Make it yours, on foundations already laid.
Open to any embedding
velella is model-agnostic: embed with Voyage, OpenAI, Mistral or Cohere, or bring vectors you already have. Whatever produced them, they meet on the same canvas, so you can explore meaning from any source.
Consult your chatbot
Add the velella MCP server and your chatbot can read the node graph you have built. Talk through the logic and the blind spots of your creation, with an assistant that can see the whole graph.
Share the whole thought
You can pass on your reasoning, as a canvas others can easily open, re-run, and push back on, or invite to a live session and think it through together, several people on one canvas in real time.
