Documentation
WOS gives an AI agent long-term memory: store what matters once, then recall only the relevant pieces per query. Memory works the same in every language: 95.2% recall@5 across 70 store-language by query-language pairs.
Three calls get you working. Create a store, write a memory into it, then recall against it. That is the whole loop:
curl -X POST https://api.wontopos.com/api/v1/memory/collection \
-H "X-API-Key: $WOS_API_KEY" -H "Content-Type: application/json" \
-d '{"user_id":"alice"}'
curl -X POST https://api.wontopos.com/api/v1/memory/store \
-H "X-API-Key: $WOS_API_KEY" -H "Content-Type: application/json" \
-d '{"user_id":"alice","content":"she prefers tea over coffee"}'
curl -X POST https://api.wontopos.com/api/v1/memory/recall \
-H "X-API-Key: $WOS_API_KEY" -H "Content-Type: application/json" \
-d '{"user_id":"alice","query":"what does alice drink?"}'Get a key in the console. A store must exist before you write to it — writing to one that does not is a 404, not a silent create.
API reference
28 endpoints, one page each — fields, limits, responses, and a curl you can paste. Start at the index.
- Writing memories Put something in, correct it later, or take it back out.
- Reading memories Semantic search, one-call context, and plain reads that return the original text.
- Images An image can be the memory itself, found by a sentence in any language.
- Stores Each end-user gets their own store. Stores are explicit: writing to one that does not exist is a 404.
- Speakers Who said each memory, so a group conversation does not collapse into one voice.
- Engrams Multi-hop retrieval pipelines the engine runs for you, with no LLM in the loop.
- Won — calls for the model Free, rate-limited calls meant for the model to ask on its own, not for your UI.
- Service What is live and what this account can reach.
The full guide
Everything in one page, written to be read start to finish: what a recall returns, what a query costs, the SDKs for Python, TypeScript, and Rust, MCP setup for Claude Code, Cursor, VS Code, Windsurf, ChatGPT and Gemini, engrams, usage tiers, and privacy. Read the guide.
- Quickstart — first call in a few minutes.
- Stores — one memory space per end-user.
- Speakers — who said what, in a group conversation.
- Images — a photo as a memory, found from a sentence.
- Engrams — multi-hop retrieval with no LLM in the loop.
- Errors & limits — statuses, rate limits, retries.
Give it to your agent
llms.txt is the whole API compressed into one file for a coding
agent to read. openapi.json is the
OpenAPI 3.1 schema. Or connect over MCP with npx -y wontopos-mcp and let the
model call memory itself.
Models and price
Tablet is the lean, low-cost model; Scroll returns a fuller context; Book is a different design still in the works. Pricing is per million tokens with no subscription — see Pricing, and the measured numbers on how we benchmark.