Documentation menu
Persistent, synthesized memory for any LLM application. Two API calls. Works with any model.
What is Anansi?
Anansi is a memory layer for AI applications. You send content to it — conversations, documents, notes, meeting transcripts — and it returns synthesized, curated context ready to inject into your LLM system prompt.
Unlike raw vector databases, Anansi runs a two-layer synthesis pass on every user's data:
- Static facts — stable knowledge about a user (up to 30). Preferences, background, recurring interests.
- Dynamic context — current state (up to 15). What they're working on now, recent friction, active projects.
- Relevant chunks — top vector search results for a specific query.
This is what you inject before an LLM call — not a wall of raw chunks, but a curated profile that fits cleanly in a system prompt.
static + dynamic come back in the exact shape a system prompt wants, so there's nothing to reshape before injection. If you'd rather have the raw scored chunks, POST /v1/search gives you those instead.Quick navigation
Quickstart
Get from zero to working memory in under 5 minutes.
API Reference
Full reference for /v1/ingest, /v1/context, and /v1/memory.
Claude chatbot
Add persistent user memory to a Claude-powered chatbot.
Voice agent
Personalise a voice agent with Anansi memory.
Multi-agent
Share memory across background and conversational agents.
Tool & action memory
Remember every user action to personalise future interactions.
Onboarding
Batch-ingest onboarding answers so the app is personalised from day one.
Notion connector
Connect your Notion workspace to the memory engine.
Meeting transcripts
Auto-ingest transcripts from any webhook source.
How it works
Create an API key
Sign up at /portal/login and generate a key. Keys start with ans_.
Ingest content
POST /v1/ingest with a userId and content string. Chunking, embedding, and synthesis queue automatically.
Synthesis runs in the background
Anansi distills accumulated content into static facts and dynamic context for each user.
Retrieve before your LLM call
GET /v1/context?userId=X&q=topic returns the synthesized profile. Inject into your system prompt.