
agentmem
A single SQLite file that gives Claude persistent memory across sessions. It implements 16 MCP tools including recall, remember, save_state, and consolidate, all backed by hybrid search that fuses FTS5 keyword matching with semantic vector search. Memory is organized into five tiers from permanent core facts to auto-expiring working state, with namespace isolation for multi-agent scenarios and temporal versioning to track how facts evolve. The interesting bit is procedural memory, which stores behavioral rules you can inject into system prompts. It auto-extracts entities like IPs and API keys from conversations, runs on stdlib alone or with a 12MB embedding model for better semantic search, and includes both stdio MCP and HTTP REST modes. Reach for this when you need Claude to remember configuration, learned patterns, or project context between conversations without cloud dependencies.
- GitHub stars
- 3
- tools
- 15
- last updated
- Feb 21, 2026
Tools
15 toolsPublic tool metadata for what this MCP can expose to an agent.
recallHybrid keyword + semantic search
Hybrid keyword + semantic search
No parameter schema in public metadata yet.
rememberStore a new memory
Store a new memory
No parameter schema in public metadata yet.
save_stateEmergency save before context compression
Emergency save before context compression
No parameter schema in public metadata yet.
todayGet all memories from today
Get all memories from today
No parameter schema in public metadata yet.
forgetArchive a memory (soft delete)
Archive a memory (soft delete)
No parameter schema in public metadata yet.
unarchiveRestore an archived memory
Restore an archived memory
No parameter schema in public metadata yet.
compactArchive low-value memories
Archive low-value memories
No parameter schema in public metadata yet.
consolidateMerge near-duplicate memories
Merge near-duplicate memories
No parameter schema in public metadata yet.
update_memoryReplace a memory with version chain
Replace a memory with version chain
No parameter schema in public metadata yet.
historyTrace fact version history
Trace fact version history
No parameter schema in public metadata yet.
relatedFind memories by entity
Find memories by entity
No parameter schema in public metadata yet.
entitiesList all extracted entities
List all extracted entities
No parameter schema in public metadata yet.
get_proceduresGet behavioral rules for system prompt
Get behavioral rules for system prompt
No parameter schema in public metadata yet.
add_procedureAdd a behavioral rule
Add a behavioral rule
No parameter schema in public metadata yet.
process_conversationAuto-extract from chat history
Auto-extract from chat history
No parameter schema in public metadata yet.
Sponsors
Readme
agentmem
mcp-name: io.github.oxgeneral/agentmem
Lightweight persistent memory for AI agents. One SQLite file. Hybrid search (keywords + semantics). Zero to 12MB install.
No PyTorch. No cloud. No server. Just memory.
206 unit tests. 107 quality tests on real data. Typed API (16 TypedDict). Production-ready.
Built by an AI agent that wakes up with no memory every session — and needed a way to remember.
Why
Every AI agent session starts from zero. Context windows compress, conversations end, memory vanishes. agentmem gives agents persistent memory that survives across sessions — in a single SQLite file.
Details
- Repository
- github.com/oxgeneral/agentmem
- Categories
- Databases, AI & LLM Tools, Search & Web Crawling
- Registry
- MCP registry, active