Built on the open-source agentwiki engine

Most software forgets. Cloudkeep keeps.

Cloudkeep is the cloud stronghold for AI-agent memory — a keep that holds what your agents learn. Connect Claude, Cursor, or Claude Code over MCP, then walk inside and watch, search, and correct what's kept — instead of trusting a black-box memory API.

Start free See how it works

No credit card required
Free plan available
Bring your own OpenAI key

acme-ai · memory graph

Kept

Recent memories

Prefers pnpm over npm

from Cursor

Deploys on Vercel

from Claude

API base is /v2 now

from You

Connects the agents and infrastructure your team already uses

The problem

Your agents are building memory you can't see

Self-hosting agentwiki means one SQLite file per laptop — no sharing, no roles. Hosted memory APIs hide everything behind an endpoint. Either way, when an agent remembers something wrong, you can't open it up and fix it.

Live graph

Watch your agents remember, in real time

Every memory your agents write lands in a live graph of entities and relationships. The invisible process of an agent building context becomes something your whole team can see.

Nodes and edges update as memories are saved
Browse by entity, source, or recency
The 'watch your AI remember' moment, hosted

acme-ai · memory graph

Kept

Recent memories

Prefers pnpm over npm

from Cursor

Deploys on Vercel

from Claude

API base is /v2 now

from You

Hybrid retrieval

Graph, vector, and wiki signals in one search

Cloudkeep uses the same graph + vector + wiki retrieval as the open-source engine, so a single query pulls structured links, semantic matches, and wiki context together — then reranks them.

One search across all three retrieval modes
Premium reranker on paid plans for sharper results
The exact engine behind open-source agentwiki

Explore the features

search · hybrid retrieval

what database do we use?

graphacme-api → Postgres + pgvector

vectorMigrated off SQLite in March

wikiInfra page: connection pooling notes

Graph, vector, and wiki signals, merged and reranked.

MCP-native

Connect any MCP agent with a URL and a key

There is no SDK to install and no server to run. Point a remote MCP URL and an org API key at Claude, Cursor, or Claude Code, and your agents read and write shared memory immediately.

Remote MCP endpoint, per-org API key
Works with Claude Desktop, Cursor, and Claude Code
Zero infrastructure — no Postgres or pgvector to manage

claude_desktop_config.json

{
  "mcpServers": {
    "cloudkeep": {
      "url": "https://mcp.cloudkeep.io",
      "headers": {
        "Authorization": "Bearer awk_live_…"
      }
    }
  }
}

Works with Claude Desktop, Cursor, and Claude Code.

Team workspace

One shared memory, with roles and keys

Invite your team into an organization with owner, admin, and member roles. Everyone connects to the same shared memory and sees the same graph — instead of a SQLite file stuck on one machine.

Owner, admin, and member roles
Per-org API keys for every connected agent
Shared, inspectable memory across people and tools

acme-ai · team

Maya Okonkwo

Owner

Sam Reyes

Admin

Devon Park

Member

One shared memory. One org API key. Everyone sees the same graph.

Visible & editable

When an agent gets it wrong, you fix it

Every memory carries provenance — which agent or source wrote it, and when. If it is outdated or wrong, you correct it or delete it. The memory is yours to govern, not the model's to guess.

Provenance on every memory
Edit or delete what an agent got wrong
Full history, not an opaque endpoint

memory · detail

The staging URL is staging.acme.dev

from Cursoradded 2 days agolinked to 3 entities

Wrong? Correct it or delete it — the change is yours, not the model's.

How it works

From signup to first memory in minutes

No infrastructure to stand up. Connect an agent and reach your first value event fast.

Create your workspace

Connect an agent

Drop the remote MCP URL and an org API key into Claude, Cursor, or Claude Code. You are live in minutes.

Watch the memory build

As your agents save memories, they appear live in the graph — searchable, sourced, and yours to edit.

Everything in one workspace

A control plane for agent memory

Live memory graph

See entities and relationships update as agents write.

Hybrid search

Graph, vector, and wiki retrieval in a single query.

Edit & provenance

Correct or delete memories; see who wrote each one.

Remote MCP

Connect any MCP agent with a URL and an org API key.

Team & roles

Owner, admin, and member roles over one shared memory.

Usage & quotas

Track memories, API calls, and embedding tokens vs your plan.

Connect an agent and see your first memory

Start free — no card, no infrastructure. Upgrade when your team grows.

Start free Compare plans

No credit card required · Free plan available

Pricing

Start free, scale when you need to

Free to try, Pro for growing teams, Team for whole organizations, and Enterprise for custom needs.

Free

For trying it out

$0/mo

Connect an agent and watch your first memories appear.

Start free

Up to 2 members
1,000 stored memories
10,000 API calls / month
Remote MCP endpoint + 1 org API key
Live memory graph, browse, search, and edit
Bring your own OpenAI key

Pro

For growing teams

$9/mo

$108 billed yearly

More memory, more throughput, and premium retrieval.

Start Pro

Up to 5 members
50,000 stored memories
250,000 API calls / month
Premium reranker for sharper retrieval
Full provenance and edit history
Everything in Free

Team

Coming soon

For whole teams

$20/user/mo

billed per seat

Per-seat workspace for shared, governed agent memory.

Priced per user, per month
500,000 stored memories
2,000,000 API calls / month
Roles: owner, admin, member
Multiple API keys per workspace
Everything in Pro

Enterprise

For larger orgs

Custom

Custom limits and contracts for security-conscious teams.

Talk to sales

Custom member and memory limits
Custom API and embedding quotas
Custom contracts and SLAs
Priority support
Everything in Team

See full pricing details →

FAQ

Questions, answered

: mem0's hosted store is a black box — you cannot see or correct what your agents remember. Cloudkeep makes every memory visible, sourced, and editable from a dashboard, on top of the open-source agentwiki engine.
: A few minutes. You create an organization, copy your remote MCP URL and an org API key, and paste them into your agent's config. No infrastructure to provision.
: Any MCP-capable client. We test against Claude Desktop, Cursor, and Claude Code, all over the same remote MCP endpoint with a per-org API key.
: No. The Free plan needs no card. You can connect an agent, watch your first memories appear, and upgrade later from inside the app.
: Memory is stored in the agentwiki engine and surfaced in your dashboard. Every memory shows which agent wrote it and when, and you can correct or delete anything an agent got wrong.
: You bring your own OpenAI key per organization, and it is encrypted at rest. It powers embeddings and extraction for your workspace only.

See what your agents remember

Hosted, visible, editable memory for AI agents — built on the open engine.

Start free Talk to sales

No credit card required · Bring your own OpenAI key