DEVOURED
ai-memory (GitHub Repo)
ai-memory provides cross-session persistence for AI coding agents by compiling project history into a git-versioned markdown wiki.
What: The open-source tool functions as a local Rust binary that bridges LLM agents like Claude Code, Codex, and Devin. It captures lifecycle events to create summary handoffs and uses SQLite for FTS5 search and entity-based recall, enabling agents to resume tasks without re-explaining context.
Why it matters: This represents a move toward 'compiled' memory architectures where agents leverage persistent, human-readable documentation rather than just relying on raw, ephemeral context windows.
Takeaway: Run `ai-memory install-mcp --client --apply` to begin capturing session history for your preferred coding agent.
Deep dive
- Persistent Memory: Compiles sanitized session observations into markdown wikis at session end.
- Cross-Agent Support: Compatible with Claude Code, Codex, Devin, Command Code, Kiro CLI, and others.
- No Vector DB Required: Uses FTS5, entity-matching, and graph-neighbor retrieval for memory recall.
- Local-First: Runs as a single Rust binary with optional Docker support, keeping sensitive project history on-premise.
- Managed Workstreams: Allows switching between agents (e.g., Claude to Codex) while maintaining a coherent session lineage.
- Autonomous Improvements: Can schedule background reviews of sessions to update the wiki with new insights.
- Auth Layers: Supports multi-user attribution and bearer token authentication for shared deployments.
Decoder
- FTS5: A SQLite extension providing full-text search capabilities.
- MCP: Model Context Protocol, an open standard for connecting AI assistants to data sources and tools.
- PreCompact: A hook triggered before the LLM consolidates logs, allowing ai-memory to capture relevant state.
Original article
Full article content is not available for inline reading.
Explore all 56 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi