OPEN SOURCE · MIT LICENSED

Your agent forgets
everything.

Xanima fixes that. Real memory with forgetting curves, knowledge graphs, and consolidation — modeled on how the human hippocampus actually works.

Formerly SharpWave. Same engine, new name. Your old brains carry over.

One-line install npx -y xanima

Vector stores aren't memory.

Your agent forgets everything the moment a session ends. The usual fix? Dump conversation history into a vector store and retrieve nearest chunks. It works until it doesn't.

It never forgets. A throwaway remark from March competes with something that actually matters.
It has no structure. Embeddings tell you what's similar. Not what caused what.
Recall degrades as it grows. More memories means more noise.

Human memory doesn't work like a vector store. It decays on a curve, strengthens what gets used, consolidates related things into concepts, and lets the rest fade. Xanima models that.

Not a metaphor. A mechanism.

📉

Real Forgetting Curves

Every memory carries FSRS-6 stability and retrievability. Unused memories decay on a power-law curve. Reviewed ones strengthen. Importance and emotional weight scale how durable a memory starts. Faded memories go dormant instead of being deleted, and come back when a search needs them.

🧠

Consolidation, Not Storage

A background pass replays recent episodes, promotes recurring patterns into durable semantic nodes, synthesizes clusters into schemas, and downscales noise. Modeled on slow-wave and REM sleep.

🕸️

A Graph, Not a Bag

Memories connect through typed edges — caused_by, supports, contradicts, supersedes, instance_of. Retrieval spreads activation across edges, so recalling one thing surfaces what's genuinely related.

🔄

Supersede, Don't Accumulate

brain_supersede closes out a stale memory and links the replacement. The graph keeps its temporal integrity instead of accumulating contradictions.

🔍

Hybrid Retrieval

Full-text search fused with vector similarity via reciprocal rank fusion, spread across the graph. Vector search is optional — FTS and graph retrieval work with no embedding provider.

🚀

Built for Fleets

One Xanima process backs every agent. Each agent's memories live in their own isolated database. A single config entry covers all of them.

What your agent gets.

brain_query Search and recall memories using hybrid FTS + vector + spreading activation
brain_write Store a new memory node. Auto-queues for embedding and auto-linking
brain_link Create a typed edge between two existing nodes
brain_supersede Replace an outdated node. Preserves temporal integrity
brain_stats Node/edge/episode counts, neuromodulator state, consolidation status
brain_review Apply FSRS-6 spaced repetition to strengthen or decay a memory
brain_forget Physically delete a node from the brain
brain_retract Mark a memory as wrong. It's kept as history but never recalled again
brain_record_decision Log a decision and what you expect to happen, then record the outcome and the lesson
brain_set_goal Set a standing goal. Sleep checks recent work against it and flags drift
brain_plan Plans with steps you can park and resume with a briefing of what changed
brain_reflect What worked, what didn't, and what to try next, built from real outcomes

Plus tools for commitments, open questions, hypotheses, briefings and more. 32 in all.

Every node is typed.

The type affects how it's consolidated and retrieved. Not everything is a fact — and Xanima knows that.

identity semantic episodic pattern skill goal emotion procedural schema decision plan commitment hypothesis reflection

One command. You're in.

OpenClaw (plugin)

openclaw plugins install npm:xanima-openclaw --force --accept-capabilities

Claude Code

claude mcp add xanima -- npx -y xanima

Any MCP Client

npx -y xanima

Memory lands in ~/.xanima/ as SQLite. Nothing leaves your machine unless you configure a remote embedding provider. Coming from SharpWave? Xanima keeps using your existing ~/.sharpwave/ brains until you move them.