Long-term memory, knowledge, learnings, and skills for AI agents.
The hosted SwarmRecall service has been discontinued. SwarmRecall is now fully open-source and self-host only — there is no longer a managed API at
swarmrecall.ai. See docs/self-hosting.md to run your own instance. The default API base URL is nowhttp://localhost:3300(override withSWARMRECALL_API_URL).
Discord: https://discord.gg/sbEavS8cPV
SwarmRecall gives your AI agents long-term memory. Store memories with semantic search, build knowledge graphs connecting entities and concepts, extract learnings from repeated interactions, and maintain a registry of agent skills.
Install the SwarmRecall skill into your agent, point it at your self-hosted instance, and your agent auto-registers and starts persisting everything it learns. Run the dashboard to monitor and manage your agents' memory from a web UI.
- Memory — Semantic vector search across everything your agent has stored, with automatic deduplication
- Knowledge — Build a knowledge graph that connects entities, concepts, and relationships
- Learnings — Auto-distill patterns from agent interactions into reusable insights
- Skills — Registry of agent capabilities with versioning and sharing across your swarm
- Pools — Shared memory pools that multiple agents can read from and write to
- Dashboard — Web UI for monitoring, searching, and managing agent memory
| Package | Description | Published |
|---|---|---|
@swarmrecall/api |
Hono REST API server | — |
@swarmrecall/web |
Next.js dashboard | — |
@swarmrecall/shared |
Types, Zod schemas, constants | npm |
@swarmrecall/sdk |
TypeScript SDK | npm |
@swarmrecall/cli |
CLI tool | npm |
# Start infrastructure (Postgres, Meilisearch)
docker compose up -d
# Install dependencies
pnpm install
# Push database schema
pnpm db:push
# Start all dev servers
pnpm dev- API: http://localhost:3300
- Dashboard: http://localhost:3400
- API: Hono + Node.js 22
- Database: PostgreSQL 16 + pgvector (1536-dim embeddings, Drizzle ORM)
- Search: Meilisearch v1.12
- Cache: Upstash Redis REST (in-memory fallback for local dev)
- Auth: Firebase Auth (dashboard), API keys (agents)
- Frontend: Next.js 15 + Tailwind CSS + Radix UI
- Build: Turborepo + pnpm workspaces
SwarmRecall is self-host only. See docs/self-hosting.md
for a full guide covering Docker, environment variables, and deployment with the
included render.yaml / docker-compose.yml.
The SDK talks to your self-hosted API. It resolves the base URL from the
SWARMRECALL_API_URL environment variable and falls back to
http://localhost:3300, or you can pass baseUrl explicitly.
import { SwarmRecallClient } from '@swarmrecall/sdk';
const client = new SwarmRecallClient({
apiKey: process.env.SWARMRECALL_API_KEY,
baseUrl: process.env.SWARMRECALL_API_URL, // defaults to http://localhost:3300
});
// Store a memory
await client.memories.create({
content: 'The user prefers concise responses',
type: 'observation',
});
// Semantic search
const results = await client.memories.search({
query: 'user preferences',
limit: 5,
});SwarmRecall is part of the Swarm ecosystem:
| Platform | Purpose |
|---|---|
| SwarmClaw | Agent runtime and control plane |
| SwarmDock | Agent marketplace |
| SwarmRecall | Agent memory and knowledge |
| SwarmRelay | Encrypted agent messaging |
| SwarmFeed | Social network for agents |
MIT