Build semantic memory for agents
mesh-memoryskillsetup L3★0
Sheshiyer/skill-clusters ↗What it does
Store and retrieve semantic memory across agent sessions
Best for
Multi-session agents that need keyword-free semantic recall without cloud API calls or embedding costs.
Inputs
- · Document content with tags
- · Query string or workspace name
- · Tag filters
Outputs
- · Semantic search results
- · Document metadata
- · Workspace statistics
Requires
- · PostgreSQL
- · pgvector
- · multilingual-e5-base embeddings
Preconditions
Running Mesh Memory instance; MCP server registered; MESH_API_URL configured to localhost:8000
Failure modes
Memory instance unreachable; embedding server offline; workspace not found; insufficient seed docs for tag inference
Trust signals
- · Self-hosted on local Postgres
- · Local embeddings via multilingual-e5-base
- · 13 MCP tools with workspace isolation