Memory

Circuit learns from what you ship. The longer you use it, the sharper it gets.

How Memory Works

Every time the team ships a spec or corrects a classification, Circuit records it. Over time, these records form a picture of how the team makes decisions: which themes get prioritised, which customer segments matter most and how specs are structured. This isn't configuration. It happens automatically as the team works.

Memory is built at the team level — it reflects collective decisions, not individual preferences. Everyone on the workspace contributes to and benefits from it.

This means memory compounds across the whole team's shipping history. If one person ships a fix for a checkout bug, the next person to work on a checkout priority sees that context in their brief — even if they weren't involved in the original ship. The team gets smarter together, not separately.

Ship Memories

When you mark a spec as shipped, Circuit writes a memory. It records:

• Theme of the shipped feature
• Volume of feedback at time of ship
• Customer segment that drove the priority

These memories accumulate. After a few ships, patterns emerge.

Classification Corrections

Correct a type on a priority and Circuit remembers. The correction applies to future feedback on the same theme. Even when new feedback arrives, the corrected type sticks.

This means less time fixing classifications and more time building. The system adapts to how the team thinks about its product.

What Specs Remember

After a few ships, specs show what Circuit remembers: previous ships, corrections and related context. If the team has shipped something similar before, the spec surfaces it. This helps avoid duplicate work and gives the builder context on past decisions.

What Circuit Knows

Go to Settings → Account to see what Circuit has learned: ship count, type breakdown (e.g. Bug 65%, Feature 23%) and any type corrections. This data is used automatically to improve specs and priority scoring.

Shipping History on Priorities

Priorities flag when they match your shipping history. If a priority aligns with themes the team has shipped before, it's marked. This gives a quick signal: has the team seen this pattern before?

Memory Lifecycle

Memories don't pile up indefinitely. Circuit applies a monthly decay rate — older memories carry less weight than recent ones. Low-relevance memories are compressed into quarterly narratives, preserving the signal without the noise. This keeps the system focused on what's current without losing long-term context.

The decay rate is 0.95 per month — a memory from three months ago carries roughly 86% of its original weight. A memory from a year ago carries around 54%. This gradual fade means Circuit stays responsive to how the team's focus shifts over time, without losing long-term context entirely.

Quarterly compression turns the lowest-weight memories into a brief narrative summary. The signal is kept; the raw event data is retired. This keeps the memory store lean and the briefs focused on what's current.

Memory in MCP

Memory enriches the MCP tools too. When pulling priorities or specs from Cursor or Claude Code:

• Priorities flag when they match your shipping history
• Specs include past context from related ships

Ask your coding tool: "Have we shipped anything like this before?" to surface memory directly.

The Compounding Effect

Each ship makes the next cycle sharper. Priorities align closer to how the team actually makes decisions. Specs reference what's been built before. Classifications get it right the first time. The loop tightens with every cycle.