Persistent Agent Memory
Portal One agents have persistent memory — durable facts that survive across conversations. Unlike conversation history (which is specific to one thread), memory contains stable information the agent should always know.
What Memory Is For
Memory stores the kind of information that reduces repetition:
- User preferences — "Jesse prefers concise responses" or "Always use TypeScript, not JavaScript"
- Corrections — "The API endpoint is /v2/users, not /v1/users"
- Environment facts — "This project uses Python 3.11 with Poetry"
- Project conventions — "We use conventional commits and squash-merge PRs"
How Memory Works
Memory is injected into every conversation turn. When an agent starts a new conversation, it automatically has access to everything it's remembered from previous interactions.
The agent actively manages its own memory:
- Adding entries — when it learns something durable about you or your setup
- Updating entries — when information changes or gets corrected
- Removing entries — when facts become outdated
Memory vs. Conversation History
| Feature | Memory | Conversation History |
|---|---|---|
| Scope | All conversations | Single conversation |
| Content | Stable facts and preferences | Full message exchange |
| Managed by | Agent (proactively) | System (automatically) |
| Size | Compact (key facts only) | Can grow very large |
Privacy & Control
Memory is scoped to your organization. You can view and manage what your agents remember through the agent settings. Memory entries are stored securely and never shared across organizations.