Memory (mem0)
Agents forget everything when a session ends. Draw an arrow from an agent to the Memory card and it can save what is worth keeping — a decision, a convention, your preference — and find it again next week, in another agent, after a restart.
Built on mem0, the open-source memory engine by mem0.ai:github.com/mem0ai/mem0
6
memory tools an agent gets with one arrow
3
scopes: this workspace, every workspace, or each agent on its own
0
accounts or servers to set up — the memory lives on your computer
Connect memory, then watch the agent use it
Draw the arrow from the agent to the Memory card yourself — or let the demo play. The tool calls and their results are the ones the real app printed in a test run.
Tools Lead has from NeuroSquad· no memory tools without an arrow
Long-term memory
0memories
No memories yet
Draw an arrow from an agent to this card — it gets memory tools and can save and recall facts across sessions.
Drag from the dot on Lead to the Memory card
Lead has no memory tools. Draw an arrow from it to the Memory card — drag from the dot, or press the button.
Claude Code in a NeuroSquad test instance with a scripted model; the tool calls, their results and the recalled text are the app’s real ones. The relevance is the built-in keyword search’s score.
Six tools, while the arrow is there
The tools reach the agent through NeuroSquad’s own MCP connection, the same way for every CLI. Remove the arrow and they are gone.
memory_searchFinds the memories relevant to a question, best first, with when and by whom each was saved.
memory_addSaves a fact — or a piece of conversation for the model to extract facts from.
memory_listLists the memories, newest first.
memory_getReads one memory by its id.
memory_updateCorrects a memory that became wrong. Its own, or any — if you allow it on the card.
memory_deleteForgets a memory. Again: its own, or any if you allow it.
Pick a scope on the card
Memories belong to the scope, not to the card: delete a card, add a new one with the same scope, and the memories are still there.
This workspace
The default. Every agent connected to a Memory card in this workspace shares one memory: the project’s decisions in one place.
All workspaces
One memory for every workspace — for what holds everywhere, like your preferences and conventions.
Per agent
Each connected agent has its own private memory in the workspace. The card shows them all, marked by agent.
Your memory, in plain sight
Everything agents save is on the card. Read it, fix it, move it or delete it — and see who wrote what.
Search and add
Search the memories the way agents do, or add one yourself with +. In the per-agent scope you pick whose memory it is.
Edit and delete
Edit a memory in place; delete it with a second press. The memory an agent has just written glows for a moment and carries its name.
History of edits
The clock on a memory opens its timeline: added, edited — with the text before — and deleted.
Export and import
Export saves the memories to a JSON file. Import shows a preview first — what is new, what is already here and will be skipped — lets you pick the scope, and Undo takes that import back.
Clear
Deletes every memory in the card’s scope, after you confirm. Export first if you may need them.
Deleting a workspace or an agent
Deleting a workspace deletes its memories; deleting an agent deletes its private ones. The delete dialog tells you how many first. Memories for all workspaces stay.
Works offline from the first minute
The card needs nothing set up. When you want more, plug in a model you already use.
By keywords, built in
Shared words and word parts: “postgres” finds “PostgreSQL”. Offline and instant, but synonyms don’t meet — “car” won’t find “automobile”. The card says which mode it is in.
By meaning: Smart search
One press downloads a small multilingual model (about 150 MB, checksums verified) that runs on your computer, offline: a question in Russian finds a memory written in English. Or use an embedding model from OpenRouter or a model server you added (Ollama, LM Studio…). Existing memories are re-indexed once.
Fact extraction, optional
Give the card a chat model and mem0 turns what agents save into short facts and skips what it already knows. One request per save.
Or let it happen on its own
Two switches on the card, both off by default. Recall reaches Claude Code, Codex, Qwen Code, OpenCode, Kilo Code, Gemini CLI, pi and omp.
Automatic recall
Off by defaultBefore each of your prompts, the most relevant memories are added to the agent’s turn — marked as notes, not instructions.
Automatic saving
Off by defaultWhen an agent finishes a turn, your prompt and its final answer go to the fact-extraction model, which keeps only lasting facts. Needs a model on the card.
One card, one arrow
Everything happens on the canvas you already have open.
Add the card
Add-card menu → “Plugin…” → Memory (mem0) → “Add to canvas”.
Draw an arrow
From an agent to the card. It gets the memory tools on its next tool listing.
Work as usual
Ask it to remember something, or let it save and search on its own.
Look at the card
Search, edit and delete the memories, see their history, export, import or clear them. The newest an agent wrote glows for a moment.
Every agent NeuroSquad runs
The tools come through NeuroSquad’s MCP connection, so every CLI on the canvas can use them.
Get the tools right away when the arrow is drawn
List their tools once at start, so they have them from the start; the arrow decides whether a call goes through
Pick them up on their next start
Stays on your computer
The memory is a few files in NeuroSquad’s data folder. Nothing goes to mem0.
Local store
Memories, their vectors and their history are stored in NeuroSquad’s data folder. No account, no cloud.
Keys stay in the app
If you pick a model, its key is only sent to that provider with each request — never written into a file or shown to an agent.
No usage reporting
The engine’s own usage reporting is switched off, and it writes nothing into your home folder.
Notes, not orders
Agents are told that memories are notes from the past that may be outdated — and never to save passwords or keys.
Open source inside
Powered by mem0
The memory engine is mem0’s open-source SDK, used unchanged under the Apache-2.0 licence and running inside NeuroSquad. The mem0 platform and its API key are not needed.
Questions about Memory
Agents that remember
Download NeuroSquad, add the Memory card, and draw an arrow.