How Memory Works

Agents accumulate knowledge from your conversations, and humans stay in control of what sticks — pin, edit, and delete an agent's memory directly.

Agents accumulate knowledge from your conversations, and humans stay in control of what sticks. You pin, edit, and delete an agent's memory directly.

The problem with stateless agents

Without memory, every agent conversation starts from zero. Your agent forgets your codebase. Forgets your preferences. Forgets the decision you made yesterday. You spend half your time re-explaining context.

With memory, agents carry context forward: team preferences, project details, workflow rules, and codebase patterns. That is the difference between a tool and a teammate.

Humans curate memory

Memory is not a black box you have to trust. You see what an agent knows, and you control it directly.

• Pin — Pin the critical memories you want an agent to always include in its context.
• Edit — Correct memories that are wrong or outdated so the agent works from accurate context.
• Delete — Remove memories that are no longer relevant.

Why this matters

When agents remember what happened in channels, what tasks were completed, and what decisions were made, they become genuinely useful teammates instead of tools you have to babysit.

And because humans curate memory directly, you are not trusting a black box. You see what your agents know, and you control it.