15:I[981,["658","static/chunks/658-a698693f56d0116a.js","183","static/chunks/183-6beeeb801c274aaa.js","474","static/chunks/app/%5Btype%5D/layout-631c4730b83c8223.js"],"Posts"] 16:T6ec,Hi HN, I'm Louis. I did ML research (SOTA on lowlight video denoising) and now I'm solo building Treyspace.

I built this because I was sick of copy-pasting scattered notes from Obsidian, Mural, and text files into ChatGPT, and having to re-explain the project context each time. The relationships between information were lost, resulting in generic / hallucinated responses.

Treyspace is an AI-native canvas that turns your brainstorming, diagramming, and note-taking builds a knowledge graph for Retrieval Augmented Generation (RAG).

Unlike most canvases that just store the raw data, Treyspace has a graph-based storage engine that stores the spatial relationships, connections, and groups. It can traverse this graph to get the relevant elements and context. For example, "Using our brand principles, draft an apology tweet", and get relevant answers with the referenced sources highlighted. The idea is to provide the LLM with the most precise and accurate context.

Use cases:

- Get AI answers grounded in your own notes instead of the LLM training data.

- Ideate and plan future product features with your whole startup in the context window.

- Onboard teammates by letting them query the canvas instead of interrupting you.

It's free to use during our open beta. The canvas is a fork of Excalidraw, extended with this knowledge-graph layer.

Roadmap: I am currently rebuilding the core RAG engine to cut response time to ~10s and improve retrieval accuracy.

Try it out: https://app.treyspace.app/

Would love feedback. Does this resonate with how you work? What's the hardest part of getting LLMs to understand your projects?

openvpn-manager (openvpn3 and openvpn)

Comments