Explain your data once. Every session after starts smarter.

As the agent works it learns what your tables mean, which joins are right, and which queries you trust. You approve each fact, and from then on every conversation starts with that knowledge.

Download for macOS Book a demo

acme / Memory

Context files

context.md

context-billing.md

context-orders.md

context-queries.md

Updated 1h ago

Billing

mrr counts active subscriptions only, trials excluded

amounts are stored in cents, divide by 100

orders joins customers on customer_id, not email

Verified query

mrr by plan

SELECT plan, sum(amount) / 100 AS mrr
FROM subscriptions WHERE status = 'active'
GROUP BY plan;

Nothing saves silently. The agent proposes, you decide.

When the agent learns something worth keeping, it asks. One click saves the fact, one dismisses it. Your memory only holds what you confirmed.

Save to memory?

mrr counts active subscriptions only, trials excluded Save

orders joins customers on customer_id, not email Saved

Memory updated context-billing.md

Plain markdown files. Read them, edit them, delete them.

Memory is not a hidden vector store. It's markdown in your workspace folder, one file per topic. A background recorder merges new facts in, dedupes, and keeps the files tidy.

context-billing.md markdown

# Billing

- mrr counts active subscriptions only,
  trials excluded
- amounts are stored in cents, divide by 100
- orders joins customers on customer_id

## Verified queries
- mrr by plan → context-queries.md

Your team learns with you. Memory syncs over git.

In a shared workspace, memory files sync like everything else. When two people learn different things, the files merge as a union, so nobody's facts get lost.

context-billing.md

+ refunds land in a separate ledger table anna

+ amounts are stored in cents, divide by 100 you

Merged as union both facts kept