Open source memory SDK for AI agents
AgentRecall gives your AI agents persistent, intelligent memory. Graph relationships, semantic search, and AI-powered processing — so every conversation builds on the last.
See how agents with memory outperform agents that forget.
🎧
The problem
Customer writes in: "We discussed this last week." Your agent has no idea. It treats every conversation like the first. Customer has to repeat themselves. Again.
With AgentRecall
Every interaction is stored as a connected memory. When the customer returns, your agent pulls up their history, preferences, and open issues — instantly.
const history = await recall.search({ query: "recent issues with this customer", agentId: "support-agent", limit: 5 });
Install the SDK, initialize with your API key, and start storing memories. That's it.
npm install agentrecall pip install agentrecall
import { AgentRecall } from "agentrecall"; const recall = new AgentRecall({ mode: "cloud", apiKey: process.env.AGENTRECALL_API_KEY }); await recall.store({ content: "User wants the dashboard redesigned with dark mode", agentId: "ui-agent" }); const memories = await recall.search({ query: "what UI changes did the user request?", agentId: "ui-agent" }); const connected = await recall.traverse(memories[0].id, { depth: 2, relationship: "relates_to" });
You don't need our cloud to use AgentRecall. Run everything on your own infrastructure with full control.
Run AgentRecall on your own server. Full control, no costs, no limits. Your data never leaves your infrastructure.
Don't want to manage infrastructure? Use our hosted API. Same features, zero setup.
Same SDK, same API, same features. The only difference is where your data lives.
AgentRecall isn't a key-value store with a fancy name. It's a memory system built for how AI agents actually need to remember.
🔗
Neo4j-powered relationship graph between memories. Memories aren't just stored — they're connected. Query by traversal, find hidden connections across your agent's entire history.
const related = await recall.traverse(memoryId, { depth: 3, relationship: "relates_to", filter: { topic: "project_alpha" } });
🧠
Qwen2.5-7B extracts entities, detects relationships, and auto-categorizes every memory. Your agent's knowledge base is enriched automatically — no manual tagging needed.
await recall.store({ content: "User prefers dark mode and React", agentId: "support-agent" });
🔍
Vector embeddings plus full-text search. Find memories by meaning, not just keywords. Your agent understands context and retrieves what's truly relevant.
const results = await recall.search({ query: "what did we discuss about pricing?", agentId: "support-agent", limit: 5, threshold: 0.7 });
☁️
RESTful API with authentication, usage tracking, and multi-agent support. Deploy in minutes, scale to millions of memories. Built for production workloads.
const recall = new AgentRecall({ mode: "cloud", apiKey: process.env.AGENTRECALL_API_KEY }); await recall.store({ content: "Session notes from standup", metadata: { team: "engineering" } });
🤖
Each agent gets isolated memory with its own namespace. Cross-agent query when needed. Perfect for teams of specialized agents working together.
const salesAgent = recall.agent("sales"); const supportAgent = recall.agent("support"); const shared = await recall.search({ query: "customer onboarding status", agents: ["sales", "support"] });
🔑
Local mode uses your own Neo4j and models. Cloud mode hosted by us. Same SDK, same API. Switch between modes with a single config change.
const local = new AgentRecall({ mode: "local", neo4j: "bolt://localhost:7687", model: "qwen2.5-7b" }); const cloud = new AgentRecall({ mode: "cloud", apiKey: process.env.AGENTRECALL_API_KEY });
One MCP server. Three platforms. Install once, remember everything.
🤖
For developers
Persistent memory for your coding sessions. Remember architectural decisions, debugging context, and code patterns across sessions.
🦞
For personal assistants
Your personal AI assistant that actually knows you. User preferences, relationships, daily routines — remembered forever.
⚡
For power users
Plug-and-play memory for CLI power users. One config line, full graph memory, semantic search. No Neo4j setup needed.
Start free. Scale when you're ready. No surprises.
$0forever
$9/month