Memory for AI agents

Give your agents
memory that lasts

Persistent, queryable memory for AI agents — retain, recall, reflect. Your agent does the work; CogniMemo owns durable storage and fast typed recall.

npm i @cognimemo/client

Trusted by 2k+ developers

api.cognimemo.com / recallLIVE
const { results } = await cm.recall({
  bank: "user@acme.com",
  query: "deployment preferences",
  types: ["preference", "procedure"],
});

// → [{ type: "preference", text: "Prefers us-east-1", score: 0.94 }]

Backed by

NVIDIA Inception ProgramCloudflareGoogle Cloud for StartupsAWS Startup Programs

Built for agent frameworks and AI tools

LangChain logo
CrewAI logo
Cursor logo
Claude logo
Windsurf logo
OpenAI logo
Anthropic logo
LangChain logo
CrewAI logo
Cursor logo
Claude logo
Windsurf logo
OpenAI logo
Anthropic logo
LangChain logo
CrewAI logo
Cursor logo
Claude logo
Windsurf logo
OpenAI logo
Anthropic logo
LangChain logo
CrewAI logo
Cursor logo
Claude logo
Windsurf logo
OpenAI logo
Anthropic logo
LangChain logo
CrewAI logo
Cursor logo
Claude logo
Windsurf logo
OpenAI logo
Anthropic logo
LangChain logo
CrewAI logo
Cursor logo
Claude logo
Windsurf logo
OpenAI logo
Anthropic logo

Memory in action

Three operations that cover the full memory lifecycle — plus integrations, self-hosting, and developer tools.

Retain

Capture what your agent learns

Store memories as text, files, or URLs. Embeddings, entities, and links are computed on ingest — so the next session starts with context, not from zero.

  • Your agent finishes a session and needs to remember user preferences for the next one.
  • Rather than stuffing everything into the context window, you call retain with typed memory blocks.
  • CogniMemo computes embeddings, entities, and links on ingest — ready for recall in the next turn.
  • Each bank is strictly isolated — keys never leak across projects or users.
See retain docs
api.cognimemo.com / retainLIVE
await cm.retain({
  bank: "user@acme.com",
  text: "Prefers concise PR summaries",
  type: "preference",
});
Our agents are finally context-aware across sessions.

Alex T.

AI Engineer

Recall

Retrieve the right context at runtime

Hybrid search fuses semantic, keyword, graph, and temporal signals — then reranks results. Filter by memory type or search everything with a single call.

  • Your agent loop needs relevant context injected before every LLM call.
  • Recall fuses semantic, keyword, graph, and temporal signals — then reranks results.
  • Filter by memory type — procedure, preference, reasoning — or search everything at once.
  • Sub-150ms latency keeps production agent loops fast.
Try the playground
api.cognimemo.com / recallLIVE
const { results } = await cm.recall({
  bank: "user@acme.com",
  query: "deployment preferences",
  types: ["preference", "procedure"],
});
MCP integration took five minutes in Cursor.

Reema K.

Startup Founder

Reflect

Synthesize insight from accumulated memory

Reflect generates disposition-aware answers from everything your agent has learned. Memories decay through lifecycle states so stale knowledge fades while frequently-used facts stay strong.

  • Your agent has hundreds of stored memories but needs a single synthesized answer.
  • You call reflect with a natural-language query instead of manually ranking recall results.
  • Lifecycle states — New, Active, Expiring, Forgotten — keep stale knowledge from polluting answers.
  • Corrections automatically reset relevant memories when users fix mistakes.
Memory model
api.cognimemo.com / reflectLIVE
const answer = await cm.reflect({
  bank: "user@acme.com",
  query: "What does this user prefer?",
});
Finally, an AI memory that actually evolves.

Samira L.

ML Engineer

Integrations

Works with the tools you already use

Native MCP for AI IDEs, SDKs for custom agents, and a gateway API for any HTTP client.

Claude logo

Claude

OpenAI logo

OpenAI

Windsurf logo

Windsurf

LangChain logo

LangChain

CrewAI logo

CrewAI

LiteLLM logo

LiteLLM

See integrations

Self-Hosting

Run the engine on your infrastructure

One container, no external services required. Same retain / recall / reflect API — your keys, your Postgres, your embedding model.

Self-hosting guide

Single container

Engine + Postgres + pgvector + ONNX embeddings bundled

BYOK encryption

AES-256-GCM at rest with your 32-byte master key

SDK-compatible

Point the client at your engine — identical API surface

Developers

Ship memory in minutes

SDKs, API, MCP server, and a live playground — pick your integration path.

Pricing

Plans that scale with your agents

Start free, upgrade when you need more quota. Pay-as-you-go available on every plan.

Free

$0/mo

Experiment and prototype with core memory operations.

  • Memories1,000
  • Recalls5,000
  • Reflects100
  • RPM60
Most popular

Pro

$29/mo

Production agents with higher quotas and room to grow.

  • Memories100,000
  • Recalls500,000
  • Reflects10,000
  • RPM600

Ultra

$49.99/mo

High-volume agents with generous monthly limits.

  • Memories500,000
  • Recalls2,500,000
  • Reflects50,000
  • RPM1,200

Scale

$99.99/mo

Unlimited memory operations for demanding production workloads.

  • MemoriesUnlimited
  • RecallsUnlimited
  • ReflectsUnlimited
  • RPM6,000
View full pricing

FAQ

Everything you need to know

Common questions about how CogniMemo works and how to get started.

CogniMemo is a persistent memory service for AI agents. It provides retain (store), recall (retrieve), and reflect (synthesize) operations — durable, typed storage that your agent reads and writes.
View all FAQs
Give your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart freeGive your agents memory that lastsStart free

Start building with memory

Get from zero to a working memory in five minutes. Sign up, create a project, and make your first retain call.

  • Retain, recall, and reflect in three API calls
  • TypeScript and Python SDKs with async support
  • MCP server for Claude, Cursor, and Windsurf
  • Self-host the full engine on your infrastructure
  • Encryption at rest with project-level policies

Persistent, queryable memory for AI agents — retain, recall, reflect.