Your agents need to remember. Today you're gluing together Pinecone + PostgreSQL + Redis + custom ETL — paying $2,000–$15,000/month and debugging sync failures at 2am. Memonic is one API that replaces all of it.
Free tier available · No credit card · MCP config in 60 seconds
Production agents need five distinct memory types. No existing database covers them all. Memonic was built from the ground up to handle all five through one API.
Current conversation context. In-memory. Request-scoped. Zero latency.
Past interactions with this user. Queryable by recency, relevance, or time range.
Learned facts and knowledge. Retrieved by meaning, not exact key. Vector-backed.
User profiles, preferences, task history. Structured. ACID. Long-lived.
What did the agent know at time T? Replay, debug, and audit past agent behavior.
Every operation your agent memory stack needs, unified in one client. No ETL. No sync. No drift.
One call stores to the right backend transparently. Vectors, relational, or cache — Memonic decides.
await memory.store({ type: 'episodic', content: 'User prefers email over SMS', userId: 'user_123', metadata: { confidence: 0.95 } })
Natural language queries across all memory types. Hybrid search with reranking. Context-window ready.
const ctx = await memory.recall({ query: 'how does this user want to be contacted?', userId: 'user_123', limit: 5, rerank: true }) // Returns ranked memories ready for LLM context injection
Query exactly what an agent knew at any past moment. Debug behavior, audit decisions, deterministic replay.
const past = await memory.rewind({ agentId: 'support-agent', userId: 'user_123', at: '2026-05-16T15:00:00Z' }) // What did the agent know yesterday at 3pm?
Production-grade RAG: ingest PDFs, docs, HTML. Automatic chunking, embedding, and indexed retrieval.
await memory.ingest({ documents: [{ content: policyDoc, source: 'policy-v2.pdf' }], namespace: 'product-knowledge' }) // Chunks, embeds, and indexes — no assembly required
Memonic exposes a first-class MCP server. Add it to Claude, Cursor, Windsurf, LangGraph, or any MCP-compatible framework — no custom wrappers, no translation layers.
Your agent gets store_memory, retrieve_memory, rewind, and create_rag_pipeline as native tools.
{
"mcpServers": {
"memonic": {
"command": "npx",
"args": ["@memonic/mcp"],
"env": {
"MEMONIC_API_KEY": "sk-mem-..."
}
}
}
}
// That's it. Your agent now has:
// store_memory, retrieve_memory,
// rewind, create_rag_pipeline
✓ Connected to Memonic in 60 seconds
No per-query charges. No vector operation fees. Flat monthly pricing that scales with your agents.
For experimenting and side projects.
For teams shipping production agents.
Full memory infrastructure for serious agent products.
Dedicated infrastructure for enterprise agent fleets.
vs. Pinecone ($70–400) + Supabase ($25–500) + Redis ($10–200) = $105–$1,100+/mo for equivalent capability
"We were running Pinecone + Postgres + Redis for our support agent. Three dashboards, three SDKs, and ETL jobs that broke every other week. Switched to Memonic. One API, everything works."
"The time-travel feature alone is worth the switch. When a user complains that the agent 'forgot' something, I can rewind to that exact moment and see exactly what it knew. Debugging agent behavior used to be impossible."
"The MCP integration dropped straight into Claude. Literally added four lines to my config and my agent could remember users between sessions. Zero custom code. This should have existed a year ago."
Pinecone, Weaviate, and Qdrant store vectors. Memonic stores agent memory — which includes vectors, but also structured state, session context, and time-versioned history. Memonic handles retrieval, ranking, and context assembly so your agent code doesn't have to.
Every write is versioned with a microsecond timestamp and transaction ID. Rewind queries replay the version log to a given timestamp, returning exactly the state that existed at that moment — no approximations, no reconstruction from logs.
Any framework that supports the Model Context Protocol: Claude Desktop, Cursor, Windsurf, LangGraph, AutoGen, CrewAI, and any custom MCP client. If it speaks MCP, it works with @memonic/mcp.
Yes. We provide migration guides for Pinecone, Weaviate, Qdrant, and PostgreSQL-based memory systems. The ingest() API can bulk-import existing embeddings in a single call.
Every memory write is tagged with agentId and userId. Retrieval is always scoped to the requesting agent — there's no cross-agent data leakage by design. Tenant isolation enforced at the API layer.
1 agent, 1GB storage, 100K operations/month, 7-day time travel, and full MCP server access. No credit card required. The free tier doesn't expire — it's a permanent tier, not a trial.
Free forever tier. MCP config in 60 seconds. No credit card.
Hobby free · Builder $29/mo · Pro $99/mo · Scale custom