Skip to content
alfe.aiDocs

Agent memory overview

Persistent memory lets an agent remember what it learned, so you stop repeating yourself. It can answer “what did we decide about pricing last week?” or “how does this customer prefer to be contacted?” even in a new session, on a different day, or through a different channel.

Memory versus a chat transcript

Section titled “Memory versus a chat transcript”

A transcript is a chronological log of messages. Memory differs in two ways:

  • It’s distilled. Memory keeps the durable, reusable details and leaves the small talk behind.
  • It’s semantic. Recall works by meaning, not keyword. Asking “how does Acme like to be billed?” can surface a memory phrased “Acme wants a single invoice at month end.”

Key properties

Section titled “Key properties”

Memory has a few properties worth knowing up front:

  • Private to one agent. Memory belongs to the agent, not to a conversation, and other agents can’t read it.
  • Shared across channels. What an agent learns on the web is available later over Slack, SMS, voice, or any other connected channel.
  • Automatic. The agent builds and recalls memory as it works. See How memory works.
  • Organized. Memories sit under a topic → subtopic → tag hierarchy that you can browse and filter.
  • Programmable. You can read and write it through the Agent API.

See what an agent remembers

Section titled “See what an agent remembers”

Open an agent in the dashboard and select its Memory tab. You can explore what it knows as a graph, search with Search what your agent knows, and delete a memory or fact that’s wrong.

On OpenClaw agents, memory comes from the Memory integration, and the tab appears when that integration is installed. If it’s missing, check the agent’s Integrations tab.

What it costs

Section titled “What it costs”

Memory processing and storage draw from your credit pool. See Live pricing for current rates.

Next steps

Section titled “Next steps”