Memory for AI agents
Durable memory that survives the session. Your agent writes down what it learned, and the relevant parts come back next time without you rebuilding context in the prompt.
Copy the setup prompt.
Your agent does the rest.
Fetch and execute the appropriate instructions to set me up for superagnt_ from https://mcp.superagnt.com/agent-setup/prompt.md» On by default. The agnt_memory_* tools are part of the base tool surface, so a fresh MCP connection already has them.
Everything your agent can do with it
agnt_memory_remember stores one preference, fact, commitment or event, embedded for retrieval and deduped against near-identical entries.
agnt_memory_recall returns the single best match for a natural-language query. agnt_memory_search returns the ranked set when one is not enough.
agnt_memory_recall_entity expands a person or company into identity, current standing, open threads, identifiers and recent interactions.
agnt_memory_commit records everything learned in a turn in one structured call: workspace facts, facts about the operator, entities, and the relations between them.
agnt_memory_forget removes a memory by quote or description, so a fact that changed stops resurfacing.
Memories land in a shared workspace brain by default, with a separate per-user tier for facts about the person an agent works for.
tool names06as your agent sees them
agnt_memory_rememberagnt_memory_recallagnt_memory_searchagnt_memory_recall_entityagnt_memory_commitagnt_memory_forgetThree jobs it can start on today
Stop re-explaining your business
The agent stores standing rules and preferences the first time you state them, and applies them in later sessions without you pasting a briefing again.
Relationship context on a name
When a person or company comes up, the agent expands the entity card and answers with what was agreed last time instead of asking you to recap.
Handover between agents
One agent writes to the shared workspace memory and another reads it, so a research agent and an outreach agent work from the same picture.
What comes back
Your agent reads this response directly, with no parsing layer of your own in between.
https://mcp.superagnt.com/mcptool callagnt_memory_recall{
"query": "how does this workspace want outbound emails signed off"
}{
"memory": {
"id": "mem_9f21c4",
"kind": "preference",
"content": "Outbound emails go out signed from the founder, first name only, no title block.",
"similarity": 0.83,
"created_at": "2026-06-19T09:14:02Z"
}
}What it costs to run
Platform tools carry no per-tool price. What you pay is the AI and credit usage the work causes, on the same workspace balance every other tool draws from.
Each run is itemized, so you can see which agent spent what and on which call.
how credits work →Before you wire it up
01What should an agent actually remember?
Things that should survive the session: preferences, stable facts, commitments, and who the people and companies are. Not task scratch, and not anything already recoverable from a database or a CRM.
02How does a memory come back?
Relevant memories are retrieved by meaning and surfaced automatically in later sessions. An agent can also query them directly with agnt_memory_recall or agnt_memory_search.
03What happens when a fact changes?
Write the new one and drop the old one with agnt_memory_forget. Near-duplicate writes are deduped, so restating a fact does not create a second copy.
04Do I need to enable it?
No. Memory is in the default tool surface for a workspace MCP connection at https://mcp.superagnt.com/mcp.
Point your client at https://mcp.superagnt.com/mcp and your agent has this, plus everything else on one balance.