Inspiration
Dev context is fragmented across GitHub, Slack, Notion/Docs, and ticketing tools. For any task, the why and the constraints live in scattered comments, threads, and docs, so coding agents like Codex and Claude Code waste time and make mistakes.
What it does
Stitch is a unified context layer for developer tools. Given a task (an issue/PR/ticket/link or a natural-language prompt), Stitch:
- Connects to your workspace tools (GitHub, Slack, Notion, etc.)
- Pulls the most relevant artifacts: threads, docs, PRs, decisions, and references
- Produces context and information across all your tools that the coding agent/LLM can use immediately
How we built it
We built Stitch with a Next.js frontend and a FastAPI backend. Supabase handles storage and ingestion. For hybrid search, we combine keyword search with semantic retrieval using embeddings from Gemini’s embedding model, orchestrated with LangGraph and LangChain, and powered by Llama 3.1 8B running on Cerebras.
Challenges we ran into
Lost keys in Uber, got food poisoning, allergic reaction, and tool integrations (Linear, Slack, etc.)
Accomplishments that we're proud of
Building a tool that solves an issue that we as developers actually experience.
What we learned
Check your pockets before you leave an Uber. And how to create an MCP.
What's next for Stitch
Supporting more integrations (Google Drive, JIRA, Teams, etc.) and image/video context processing.
Built With
- cerebras
- fastapi
- gemini
- langchain
- langgraph
- next.js
- supabase
Log in or sign up for Devpost to join the conversation.