Everything you need to govern agent memory
Cloudkeep turns the open-source memory engine into a hosted, team-shared control plane — visible, editable, and connected over MCP.
See and shape what your agents remember
The memory your agents build is visible, sourced, and yours to edit.
Live memory graph
A graph of entities and relationships that updates as agents save memories, so you can watch context form in real time.
Edit and correct
Fix or delete anything an agent got wrong. The memory is governed by your team, not guessed by the model.
Provenance
Every memory records which agent or source wrote it and when, so you always know where a fact came from.
Hybrid retrieval from the open engine
The same graph + vector + wiki retrieval as open-source agentwiki, hosted.
Hybrid search
One query spans graph links, semantic vector matches, and wiki context, then merges the results.
Premium reranker
Paid plans add a reranking pass that sharpens which memories surface first for a query.
Entities and relationships
Memories link into entities so retrieval understands how the things your agents track relate.
Connect agents with no infrastructure
A remote MCP endpoint and an org API key — no SDK, no server to run.
Remote MCP endpoint
Point any MCP-capable agent at one URL with an org API key and it reads and writes shared memory immediately.
Works with your agents
Tested with Claude Desktop, Cursor, and Claude Code over the same remote endpoint.
Zero infrastructure
No Postgres, no pgvector, no MCP server to operate. The hosted engine handles storage and retrieval.
Shared memory, with roles and limits
An organization workspace with roles, keys, and usage you can see.
Roles and members
Owner, admin, and member roles over one shared memory, instead of a SQLite file stuck on a single laptop.
Org API keys
Issue per-org API keys for connected agents. Keys are shown once and stored hashed.
Usage and quotas
Track memories, API calls, and embedding tokens against your plan from the dashboard.
Open a memory, see where it came from, fix it
Unlike a black-box memory API, every memory in Cloudkeep is something you can inspect. See the source agent and timestamp, follow its entity links, and correct it when it drifts.
- Source and timestamp on every memory
- Edit or delete in a click
- Entity links show how memories connect
The staging URL is staging.acme.dev
Wrong? Correct it or delete it — the change is yours, not the model's.
A URL and a key is the whole integration
Connecting an agent is configuration, not code. Paste the remote MCP URL and an org API key into your agent and it starts reading and writing shared memory.
- Remote MCP endpoint, per-org API key
- Claude Desktop, Cursor, and Claude Code
- No SDK and no infrastructure to manage
{
"mcpServers": {
"cloudkeep": {
"url": "https://mcp.cloudkeep.io",
"headers": {
"Authorization": "Bearer awk_live_…"
}
}
}
}Put your agents' memory in the open
Start free and connect your first agent in minutes.
No credit card required · Free plan available