The Problem
AI conversations suffer from context loss. Each new chat starts with "let me give you some context" as critical decisions, preferences, and insights disappear between sessions. Existing tools create silos—memory doesn't persist across applications, and associations between ideas go unrecorded. This forces users to re-establish context repeatedly, wasting mental cycles on information already provided.
What This Does
CORE positions itself as a "digital brain" that organizes episodes into topics and creates associations between them. The codebase is a React/TypeScript application with 186 files in apps/webapp/app/, featuring a conversation system (apps/webapp/app/components/conversation/), bullMQ queue/worker infrastructure (apps/webapp/app/bullmq/), and graph visualization components (apps/webapp/app/components/graph/). Queue management lives at apps/webapp/app/bullmq/queues/index.ts and apps/webapp/app/bullmq/workers/index.ts, suggesting background job processing for memory operations. The graph module includes clustering visualization and filtering capabilities, indicating a topological approach to memory organization. Integration points exist for API keys, OAuth, and MCP (Model Context Protocol) authentication, positioning this as middleware between AI tools and memory storage.
How To Use It
Setup: No root-level package.json is visible in the file listing, but the React setup implies npm or pnpm dependencies within apps/webapp/. The .env.example file at the root and apps/webapp/.env.example (implied by the .env.example presence) would configure environment variables.
Configuration: Environment variables likely reside in .env.example at the root level and apps/webapp/.env.example. The .github/workflows/build-docker-image.yml suggests Docker-based deployment is supported.
Running it: The Docker workflow indicates docker build and docker run are the primary deployment mechanisms. No CLI entry point or main.ts is evident in the file structure—usage appears to be through the web application at apps/webapp/app/.
Real-World Use
A developer working across Cursor, Claude, and other AI tools could integrate CORE as a persistent memory layer. Instead of each session starting fresh, the AI would query CORE's topic-organized memory to surface relevant prior decisions, code patterns, or project context. The bullMQ queue system would handle async memory operations—indexing new conversations, updating associations, or fetching context—without blocking the AI response cycle. The graph visualization components (apps/webapp/app/components/graph/graph.tsx, apps/webapp/app/components/graph/graph-clustering.tsx) could help users audit what memories are active and how concepts connect.
Code Health & Issues
[Medium/SDLC] No test files detected - repository-wide. No tests directories, no test configuration beyond ESLint/Prettier. This means memory logic, queue processing, and API routes have no automated verification. [Medium] No root-level package manager configuration - the root has .npmrc but no package.json at the top level. Dependency management and install commands are ambiguous without exploring apps/webapp/package.json. [Low] GitHub Actions workflows present (.github/workflows/build-docker-image.yml, .github/workflows/submit.yml) but no evidence of test execution or linting in CI. Linting exists (apps/webapp/.eslintrc, apps/webapp/.prettierrc) but may not run on pull requests. [Low] License file present (LICENSE) but no SECURITY.md policies beyond the standalone SECURITY.md at root—typical for open-source projects but worth verifying contribution guidelines.
The Bottom Line
This is a functional React memory-agent framework with bullMQ integration and graph visualization, suitable for developers who need persistent AI context across sessions. The absence of tests and ambiguous root-level packaging are the primary barriers to production use. Teams comfortable with Docker-based deployment and TypeScript React patterns can extract value quickly, but the untested codepaths and missing dependency clarity at the repo root warrant vetting before commitment.