How I Built an AI Agent That Uses a Graph Database as Its Memory

Description and use

ContriGraph uses FalkorDB to recommend open source issues, recording each interaction as a typed edge, creating persistent memory without session state.

Problem

Finding good open source issues to contribute to is harder than it looks. The real problem is memory and context, not discovery.

Outcome

ContriGraph delivers issue recommendations in under 1 ms per query, filters out previously skipped or applied issues at the database level, and uses graph edges as persistent memory, eliminating the need for session state or separate action tables.

Workflow

• User types a query in the browser
• FastAPI streaming endpoint receives the query
• Claude AI agent decides which tool to call
• Tool executes a Cypher query against FalkorDB
• Result streams back via SSE to the browser
• Agent writes VIEWED, BOOKMARKED, SKIPPED, or APPLIED_TO edges to FalkorDB
• Next query uses NOT clauses to filter out previously interacted issues

Evidence

Finding good open source issues to contribute to is harder than it looks.

Source

ContriGraph is an AI agent that traverses this graph to find the right issue for you — and remembers every interaction by writing it back to the graph as a typed edge.

Source

When you click Bookmark, Skip, or Apply on a card: CREATE (d)-[:BOOKMARKED {timestamp:$ts, session_id:$session}]->(i)

Source