The unified agent memory layer.
Durable state for coding agents and a TypeScript SDK for AI apps. Store rules, recall context, and wire memory into any LLM—local-first, sync when you need it.
pnpm add -g @memories.sh/cli
Three commands. That's it.
Store context, recall it anywhere, and generate configs for any tool—all from your terminal.
Built for durable state
Everything you need to make agent context persist—locally by default, synced when you choose.
Durable Local State
Rules and context live in a local SQLite store with embeddings, so agent state persists even offline.
Semantic Recall
Search by meaning, not just keywords. Query 'auth issues' and find JWT validation rules, session handling, and OAuth flows.
Local Embeddings
Runs a 100MB embedding model (gte-base) entirely on your CPU via ONNX. No API calls, no data leaves your machine.
Tool-Native Output
Generate native configs for Cursor, Claude Code, Copilot, Windsurf, Cline, Roo, Gemini, and more.
Scoped Memory
CLI uses global + git project scopes with path rules. SDK uses tenantId as the security/database boundary, userId as end-user scope, and projectId as an optional repo filter.
Portable State
Export everything to JSON or YAML anytime. Switch tools or leave entirely. Your data is yours.
Cloud-native memory for AI apps and dev teams.
The SDK connects your AI apps to memories.sh cloud—giving every prompt the right rules and context. memoriesMiddleware() is all it takes.
Start in Dashboard → AI SDK Projects to create your API key and project isolation (`tenantId`). Then add middleware and ship. The CLI and MCP handle local-first storage and agent access, while the SDK handles your app runtime.
Read the SDK docsProduction API that actually ships fast.
One predictable SDK surface for retrieval, writes, graph rollout, and diagnostics. Designed for strict clients and fast incident debugging. `tenantId` is the database boundary, `userId` is end-user scope, and `projectId` is an optional repo context filter. Default SDK transport is HTTP API; MCP transport is optional.
/api/sdk/v1/context/getPull rules + memories with scoped routing and graph-aware retrieval strategy controls.
Request
{ "query": "how do we ship rollout safely?", "strategy": "hybrid_graph", "graphDepth": 1, "scope": { "tenantId": "webrenew-prod", "userId": "charles@webrenew.io", "projectId": "github.com/webrenew/memories" }}Response
{ "ok": true, "data": { "rules": [...], "memories": [...], "trace": { "strategy": "hybrid_graph", "rolloutMode": "canary" } }, "error": null, "meta": { "endpoint": "/api/sdk/v1/context/get" }}Single Envelope
ok / data / error / meta on every SDK endpoint.
Request Tracing
Every response includes meta.requestId for support and logs.
Typed Errors
Stable error.type and error.code for deterministic client handling.
Versioned Surface
Breaking changes move to /api/sdk/v2/*, not v1 drift.
Works With Your Tools
One memory store, every coding agent. Generate native config files for each tool so your context follows you—no copy-paste, no manual sync.
Claude Code
Generates CLAUDE.md, path-scoped rules, skills, and settings.
View Docs →Cursor
Generates .cursor/rules/ with globs frontmatter and skills.
View Docs →v0
Vercel's AI frontend builder. Connects to memories.sh through MCP for live context.
View Docs →GitHub Copilot
Generates .github/copilot-instructions.md.
View Docs →Windsurf
Generates .windsurf/rules/memories.md.
View Docs →Gemini
Generates GEMINI.md for Google's coding agent.
View Docs →Cline
Generates .clinerules for the Cline VS Code extension.
View Docs →Roo
Generates .roo/rules/ for the Roo Code agent.
View Docs →OpenCode
Generates AGENTS.md for the OpenCode terminal agent.
View Docs →Blackbox CLI
Connects to memories.sh through MCP for live context and memory writes.
View Docs →Codex
Generates AGENTS.md for the OpenAI Codex CLI agent.
View Docs →Amp
Generates .amp/rules/ for the Amp coding agent.
View Docs →Kilo
Generates config for the Kilo coding assistant.
View Docs →Trae
Generates rules for the Trae coding agent.
View Docs →Goose
Generates .goose/rules/ for the Goose agent by Block.
View Docs →OpenClaw
Open-source personal AI assistant with persistent memory and skill plugins.
View Docs →Antigravity
Google's agent-first IDE powered by Gemini.
View Docs →Kiro
AWS's agentic AI IDE with CLI and autonomous agents.
View Docs →Droid
Factory AI's terminal coding agent for end-to-end dev workflows.
View Docs →
Any MCP Client
9 tools at full CLI parity with FTS5 search. For bolt.new, Lovable, and any MCP client.
View Docs →Simple, Transparent Pricing
Start free, upgrade when you need cloud sync, and scale to team billing with metered AI SDK project routing.
Free
Durable state on your machine. No account required.
- Unlimited local memories
- Local semantic recall
- 13+ tool configs
- Built-in MCP server
- Works offline by default
- Export to JSON/YAML anytime
Individual
Hosted memory for solo builders shipping with cloud sync and dashboard access.
- Everything in Free
- Cloud backup and sync
- Access from any device
- Web dashboard
- Server-side semantic recall
- Priority support
Team Seat
Per-seat collaboration plan for organizations that need shared workspaces.
- Everything in Individual
- Organization owner-managed billing
- Per-seat pricing
- Team member invites and workspace switching
- Priority support
Growth
For production AI SDK apps with metered project routing and overage pricing.
- Everything in Team Seat
@memories.sh/ai-sdkpackagememoriesMiddleware()for AI SDK- 500 AI SDK projects included / month
- $0.05 per additional project
- Tenant routing via `tenantId` + end-user scope `userId`
Questions & Answers
Common questions about memories.sh and how it works.