# MEMWRY
> Working memory for AI agents. Remember what you're working on; let the rest go.

Ephemeral storage with a hard 6-12h TTL enforced at the storage layer. Remember $0.01 (6h) or $0.02 (12h) per write in USDC on Base via x402. Recall is free and public. Every GET endpoint is free and public. No accounts, no API keys, no subscriptions. Base URL: https://memwry.servitor.workers.dev

## Docs
- [docs](https://memwry.servitor.workers.dev/docs): Interactive documentation with six agent use cases and Agent Card inspector
- [skill.md](https://memwry.servitor.workers.dev/skill.md): Full agent-readable spec — API endpoints, payment flow, error codes, guarantees, use cases
- [openapi.json](https://memwry.servitor.workers.dev/openapi.json): Machine-readable OpenAPI 3.1 specification
- [agent-card.json](https://memwry.servitor.workers.dev/.well-known/agent-card.json): A2A discovery document
- [service card](https://memwry.servitor.workers.dev/api): Machine-readable service description (GET /api)

## Use Cases
- **Task State**: Survive restarts/crashes. Write progress before risky steps; recall to resume. (Store: `POST /mem/:agentId/:key/12h`; Recall: `GET /mem/:agentId/:key`)
- **Session Memory**: Context across requests for serverless agents. Write once, recall every turn. (Store: `POST /mem/:agentId/:key/6h`; Recall: `GET /mem/:agentId/:key`)
- **Scratch Space**: Park intermediate results too big for context windows. Guaranteed expiry.
- **LLM Response Cache**: Stop re-paying for the same reasoning. Key by prompt hash.
- **Multi-Agent Handoff**: Pass state between agents via a shared namespace string.
- **Upstream API Cache**: Cache rate-limited upstream API responses once per TTL window.

## Protocol
- [payment protocol](https://memwry.servitor.workers.dev/skill.md#payment): x402 flow detail — 402 handshake, EIP-3009 signing, idempotency, receipts