Evolution memory

Evolution memory is where the improvement flow deposits confirmed and shipped patterns. Not a space you or the Agent write into: it’s read-mostly, and Agents that attach it benefit passively. What lands here:

• Diagnostic patterns failure → confirmed cause → winning fix chains
• Efficacy scores how often each pattern’s reuse actually holds up
• Freshness external facts decay if nobody reprobes them, surfacing revalidation candidates

​

When to use it

• Long-lived Agents where quality should compound
• High-volume Agents where the improvement flow accumulates patterns quickly
• Teams that want visibility into how the diagnostic memory grows

​

Creating one

Memory → New space → Evolution (self-evolving) creates one immediately. You don’t fill it directly; the improvement flow does, one entry per confirmed-and-shipped improvement.

​

The detail page

Evolution has two sub-tabs inside its primary lens:

• Metrics dashboard showing total entries, reuse/recurrence counts, efficacy and freshness distributions, growth curves over time
• Graph the diagnostic patterns this space has accumulated, rendered as a 3D graph. Click and drag nodes to explore
• Log events in chronological order
• Settings covered below

​

What you configure in Settings

Evolution is auto-managed, so there’s not much to tune. Basics rename the space. Flow binding bind to a specific Flow to scope patterns to that Flow’s Agents. Leave unbound and attach per-Agent. Scope workspace / tenant partitioning (see Scoping). Access (optional) partition key + required-permissions gate. No policy prompt or retrieval-tuning fields here. Filling, decaying, and revalidating are handled by the improvement flow and the canary system.

​

Attaching to an Agent

Flow builder → Agent → Memory tab → Add memory. Once attached, the Agent recalls relevant patterns automatically during diagnosis.

​

Pruning nodes

In the Graph tab, click a node and hit Delete to soft-remove that diagnostic entry. Use this to strip out patterns you don’t think are actually valid. Right after removal, an Undo toast appears in the bottom-right. Click to reverse. It’s a soft remove, so admin tools can recover it later too.

​

Relationship with the improvement flow

Evolution is where improvement-flow results land. Every loop through signal → cluster → improvement proposal → simulation → approval → deploy leaves a new diagnostic pattern here. See Improvement overview for the full picture.