Memory
Your agents remember what they learn. As they work with you, they build up long-term memory about your business, your people, and how you like things done. Ask a follow-up next week and the agent can recall what you told it today.
How memory works
- They learn as they go. After conversations and work, Cohort saves the facts worth keeping to long-term memory. Agents can also save something on purpose when you ask them to remember it.
- They look things up when it matters. When a request touches something from the past, the agent searches its memory and uses what it finds.
- Each agent has core memory. Alongside long-term memory, each agent keeps a short core memory about who it’s working with and how. It reads this at the start of every session.
- Memory is shared across your workspace. Anything a member tells any agent can be recalled by any agent in the workspace, and shared with other members of the workspace. That includes private chats and texts. Don’t tell an agent anything you wouldn’t share with your team. Before an agent asks you for personal context in a first conversation, it reminds you of this.
The Memories page
Open Memories in the sidebar to see everything your agents have learned.
- List (the default) shows memories newest first. Switch the layout to 2D or 3D to see the graph of people, things, and how they connect.
- Filter by tier (core or long-term) and by fact type.
- Open a memory to see where it came from, when it was last used, and how many times it has been confirmed.
- Edit a memory’s text by clicking it, delete it, or restore one you deleted.
- New Memory saves something you want your agents to know, with optional context such as where it came from or when it applies. It appears after the agent’s next memory sync.
Each agent’s core memory shows as its own entry. You can read it on the Memories page; it isn’t edited there.
Turning memory off, or erasing it
Settings → Memory has two controls:
| Control | What it does |
|---|---|
| Long term agent memory | Turn it off and agents stop building and using long-term memory. Turn it back on any time. |
| Erase all memory | Permanently deletes everything your agents have learned about this workspace. This can’t be undone. |
Long conversations
A conversation can go on for a long time. When one gets very long, Cohort summarizes the earlier part so the conversation can keep going. The agent keeps working from that summary, and anything it saved to memory is still there to recall.
From the API
Read memories with the Memories API (GET /api/v1/memories).
Last updated on