shashanknidhi/aznex

Aznex is a tool that lets a team of developers share the knowledge their AI coding assistants build up while working on a shared codebase. Tools like Claude Code or similar coding agents pick up useful context as they work, such as odd build quirks, why a certain design decision was made, or dead ends already tried, but normally that knowledge stays trapped on one person's computer and tied to one specific tool. Aznex instead stores it centrally, scoped to a single repository, and makes it available to any coding agent that speaks a shared connection standard called MCP. The system works through small hook adapters installed on each developer's machine that watch coding-agent sessions and send durable pieces of knowledge to a background worker. That worker cleans out anything that looks like a password or secret, twice, before sending it to a central service. The central service is the only part that touches the database, and it is also responsible for checking that a user actually has permission to access a given repository through the connected code host. It also flags older stored knowledge as possibly outdated if the code it refers to has since changed. The project is split into four pieces: shared type definitions, the local background worker that captures and scrubs data, the central service that stores data and answers agent queries, and a read-only web app for browsing and searching what the team has learned. It is built with TypeScript running on Bun, uses a small web framework called Hono for the service, SQLite moving toward Postgres for storage, and reuses a developer's own Claude Code login rather than requiring a separate API key. Getting started for an individual developer is a single install command supplied by a team administrator, which signs the developer in through GitHub and quietly sets up the background worker and hooks. Setting the system up for a whole team requires an administrator to deploy the central service, register a GitHub App and OAuth app, and choose which repositories are onboarded. The project is still in early development and is licensed under Apache 2.0.