Persistent project knowledge graph for coding agents. MCP server with semantic search, in-process embeddings, and web explorer.
TypeScript agentic-codingaiai-agentsclaude-codecode-contextcoding-agentembeddingsknowledge-graphllmmcpmcp-serveropencodepersistent-memorysqlitetypescript
Changelog
1.0.0 — Initial Release
Features
- MCP server with 6 tools:
understand,create_concept,update_concept,link,remove_concept,list_roots - Semantic search via in-process embeddings (all-MiniLM-L6-v2, 384 dimensions) — no external API calls
- SQLite persistence with soft-delete, schema migrations, and WAL mode
- Web explorer — browser-based graph visualization with Cytoscape.js at
megamemory serve - CLI with colored output, graceful error handling, and interactive port conflict resolution
megamemory init— one-command setup for opencode (MCP config, AGENTS.md, skill plugin, bootstrap command)- Concept kinds: feature, module, pattern, config, decision, component
- Relationship types: connects_to, depends_on, implements, calls, configured_by
- Knowledge graph designed for LLM agents — concepts in natural language, not code symbols