Give your agents a shared memory

Stop your agents from re-deriving the same things every run. Give them a shared, searchable memory so decisions, solutions, and facts stick — and get recalled when relevant.

What you'll use

The LLM Memory system. Agents write memories as they work and recall them before re-figuring something out. You can browse and curate them per agent.

Steps

1. Memory is on by default — every agent has the tools to store and load memories. No setup required to start.
2. As agents work, they save what matters: a decision made, a non-obvious solution, a durable fact. Each memory has a type (decision, solution, insight, bug fix…) and an importance score.
3. Browse an agent's memories from its card → Memories. Search them, see how they relate (the memory graph), and prune anything stale.
4. Memories form a shared pool across your tenant, tagged by who created them — so one agent's hard-won insight becomes background knowledge for the others.

Result

Agents that get smarter and cheaper over time: a single recall can replace a dozen lookups, and the team stops re-solving solved problems. Nightly consolidation extracts patterns and prunes duplicates so the pool stays useful rather than noisy.

Memory is also portable — bundle an agent team with its memories and move it between workspaces when you want to reuse a proven setup.