Skip to content

User Guide

What you will find here

Topic When you need it
Embedders Pick the right vector model (SentenceTransformers, OpenAI, ONNX int8).
In-Process Library Use Semvec as a library without the REST API server; bring your own embedder.
Integrations LangChain, DeepAgents, PostgreSQL, Neo4j, mem0.
Correcting memories Override wrong facts; meta_filter, per-memory provenance, anti-resonance attractors.
Detecting updates (DedupSignal) RAG / agent frontends: read-only "is this an update of something we already have?" hint. Deterministic, no LLM call.
Coding agents (overview) Persistent memory for Claude Code, Cursor, custom MCP clients.
· Claude Code MCP server + SessionStart / PreCompact hooks.
· Cursor MCP server + semvec.mdc project rule.
Cortex (multi-agent) In-process SemvecAgentNetwork, SemvecCortexService, or REST.
Cortex over REST API Clusters, regions, observers, network endpoints.
Compliance pack Event store, retention, deletion certificates, HMAC, RS256.
Troubleshooting Symptom-driven fix table.
FAQ "When to use Semvec vs …", licensing edge cases, offline use.

When to use what

  • Default starting point: semvec serve (REST API). Lowest setup friction, polyglot, fixed endpoint shapes.
  • Multi-agent coordination: start with Cortex over REST (/v1/cluster/*, /v1/region/*) — same low setup friction as the base REST API, distributed-ready out of the box.
  • Tighter per-turn latency / in-process state: drop into the in-process library — SemvecState + SemvecChatProxy.
  • Multi-agent inside one Python process: Cortex in-process (SemvecAgentNetwork) once you are already on the library.
  • Regulated workload: add the compliance pack on top of any path above.
  • Coding agent (Claude Code, Cursor): use the coding-agents MCP server.

For a side-by-side decision tree, see Choose your path.

See also