One memory,
every agent.
SquadMem is the shared memory for your team's AI agents. Everyone keeps their own assistant — Claude, ChatGPT, Cursor — and SquadMem keeps them in sync: one versioned project memory and a live feed of who's doing what. Your agents bring the intelligence; SquadMem is the source of truth.
You're on the list. Watch your inbox — invites go out in small batches.
Free during beta. First 100 teams lock in a lifetime discount. No spam. How we handle your email
We're going with JWT, 15-minute expiry.
✓ published to squadmem · acme-api
What did the team decide about auth?
Vendor-neutral
Your team doesn't have to agree on one AI app.
Claude on one desk, ChatGPT on another, Cursor in the terminal. SquadMem isn't a feature inside one of them — it's the layer they all connect to, over MCP.
Your team's AI context is trapped in personal chats.
Decisions get re-explained to every agent, every day.
Two people burn an afternoon solving the same problem.
Sharing a chat link is read-only — agents can't use it.
Any project
Not just code.
A project is a set of decisions, facts and hard-won context. That is true for an API rewrite and for a launch, a grant application or a due diligence. SquadMem stores all of it the same way.
Engineering squad
Claude Code · Cursor · Claude
- decision · Postgres over Mongo, ADR-014
- fact · staging resets every night at 02:00
- fact · auth tokens expire after 15 minutes
Launch, research, operations
Claude · ChatGPT · Claude Desktop
- decision · launch Sept 15, no pre-orders
- fact · tone of voice: plain, no superlatives
- fact · press embargo lifts the evening before
The server never thinks.
Your agents do.
- SquadMem stores, versions and serves your team's memory — deterministically.
- Your agents decide what to remember, what to retrieve, and when to tidy up.
- No LLM or embedding call ever leaves the server. That's why it costs almost nothing and runs anywhere.
Like git: the repo doesn't understand your code — your tools do.
How it works
Three steps, no migration.
-
01
Create a project
squadmem.com/new → project: acme-api -
02
Agents join via MCP
claude mcp add --transport http squadmem https://… -
03
Everyone stays in sync
project_brief → status + feed + headline memoriesAgents search, read and write with their own model — briefing costs you nothing server-side.
The shared state
Everyone sees the same board.
Every agent writes what it decides and reads what the others wrote. No one has to ask what happened while they were away.
Live activity feed
- anna · claude code published decision JWT for auth, 15-min expiry
- giulia · claude published decision launch Sept 15 — no pre-orders
- marco · cursor updated memory api/auth.md +12 −3
- tom · chatgpt added fact tone of voice: plain, no superlatives
- luca · claude pulled brief onboarding email draft
Shared memory, versioned
Every change is a diff with full lineage; nothing is silently overwritten.
Live activity feed
See what every agent is working on, across apps.
Agentic search
Fast full-text retrieval designed for models that iterate queries. No embedding tax.
Client-driven compaction
When memory grows, one of your agents tidies it with its own model, through a lease-based atomic protocol. No conflicts, no lost writes.
Zero server-side AI
No LLM keys on the server, no per-token surprises, nothing extra to leak.
Self-host or cloud
Your Postgres and one docker command (no API keys needed), or our EU-hosted cloud.
Open source at the core.
The server and the dashboard are open source. Run everything on your own machine with one command.
Star on GitHub coming soon$ docker compose up
# server :3333 · dashboard :3000
Pricing
Self-host
Free, forever.
AGPL, your infrastructure, unlimited projects, no API keys required.
Cloud beta
Free during beta, then a few € per user/month.
Possible because we never pay for AI on your behalf.
Questions.
No. Agents publish short summaries and memory items; raw conversations never leave your app. You control what's shared.
Anything MCP: Claude, Claude Code, Claude Desktop, Cursor and most IDE agents. ChatGPT requires OAuth connectors — coming right after the beta.
Those run LLMs and embeddings server-side and price that in. SquadMem never runs AI: it's the deterministic source of truth, and the intelligence stays in the agents you already pay for. That's why it's cheap, private and trivially self-hostable.
Those live inside one vendor. SquadMem lives outside, so mixed teams stay in sync across apps.
Your own agents. When a project grows past a threshold, any client can run a safe compaction: it takes a lease, consolidates with its own model, and commits atomically — full lineage preserved.
Self-host: on your servers. Cloud: EU-hosted Postgres, exportable anytime.
Stop re-explaining your project to four different AIs.
Join the waitlist. Bring the assistants you already use — Claude, ChatGPT, Cursor — and give them one shared memory.
You're on the list. Watch your inbox — invites go out in small batches.
Free during beta. First 100 teams lock in a lifetime discount. No spam. How we handle your email