Session Indexer

session-indexer

Per-project semantic search over Claude Code session history. Indexes JSONL
transcripts into a per-project SQLite store; retrieves via bge-m3 embeddings
(Ollama) with FTS5 BM25 fallback. Automatically injects relevant past context
at session start.

Problem it solves: returning to a project after a week and needing to find
"what did we decide about X" across dozens of past sessions. session-end
gives you "where I left off last time"; session-indexer gives you "what we
discussed across all history" — by semantic similarity, not grep.

Scope: single developer, single machine. This indexes *your own*
individual sessions with Claude Code in a project — not a team's shared
history, not a multi-user store. If you need to share findings with
teammates, that's a conversation/PR/doc, not something this tool does.
See below for why that's a deliberate choice, not a limitation.

Why not a centralised memory tool?
mempalace,
agentmemory, and
MemMachine all run on a single
shared backend — mempalace in ChromaDB, agentmemory via an iii engine MCP
server, MemMachine via a Neo4j + SQL backend behind a REST server
(self-hosted or their managed cloud; MemMachine does add logical per-tenant
isolation via org/project IDs, unlike the other two). That single-backend
architecture still has one fatal flaw: if it dies, everything on it dies
at once. A corrupt ChromaDB index, a crashed MCP server, or an unreachable
MemMachine/Neo4j instance takes down memory for every project and tenant
depending on that instance simultaneously, and recovery is non-trivial.
MemMachine in particular targets multi-tenant SaaS agent products (CRM,
healthcare, finance assistants) — a different problem than a solo dev's
per-project recall tool.

session-indexer is per-project and append-only…