Skip to content

github-pr-memory

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

github-pr-memory

README.md

GitHub PR memory using Exabase

This is a simple example of ingesting public pull-request discussion and review comments into Exabase memories, then chatting with that context using Exabase, Next.js, tRPC, AI SDK, shadcn/ui, and Tailwind CSS.

You create an isolated Exabase Base (your baseId is in the URL). Paste a public github.com repository URL, then Ingest & synthesize from PRs to load recent issue-thread and file-level review text from up to five of the most recently updated pull requests. An OpenAI model (via the AI SDK) reads that text and only creates memories for high-signal, durable team knowledge—review themes, conventions, things to avoid, and similar—rather than one memory per raw comment. The chat can search and add more memories. Populate sample PR memories is a no-GitHub demo pack.

How Exabase is used in this example

  • Create a dedicated Base per demo session so everything is scoped under /b/[baseId].
  • create memories by combining the GitHub REST API (fetch) with a structured generateObject call that filters and condenses review discussion into a small set of plain-text memories (infer: false on Exabase create).
  • search and add from the chat so the assistant can reason over that distilled context and store new follow-ups.

Demo

demo-x2.mp4

Deploy your own

Deploy with Vercel

Add GITHUB_TOKEN in the Vercel project settings if you hit unauthenticated GitHub rate limits; public data does not require a token but limits are low.

Technologies

  • Exabase: Bases for scoped isolation, batch memory creation, search, and writes
  • Next.js: Front-end and back-end
  • tRPC: Type-safe APIs for memory list, sample seed, GitHub ingest, and new-base creation
  • GitHub REST API: Public repository pulls, issues comments, and pull review comments
  • AI SDK: Structured synthesis (generateObject) for ingest, plus streaming chat and tool calling (search and add memory)
  • shadcn/ui: Accessible UI components
  • Tailwind CSS: Styling
  • Biome: Lint and format

Run locally

bun install
bun run dev

Create a .env.local file with EXABASE_API_KEY and OPENAI_API_KEY. Optional: OPENAI_MODEL, GITHUB_TOKEN (see .env.example).

Open http://localhost:3000 and click New base; you are redirected to /b/<baseId> where the app runs.