Pricing

Two meters. Nothing else.

You pay OpenAI for your model calls directly, with your own key. We meter the memory: the turns it serves and the history it holds.

meter 1 · per user turn

User turns

Price to be announced

A turn is a decorate call that carries a new user message. It covers everything memory does for that turn and the tool loop that follows it.

  • Recall for the new message: searches, the walk, distill after the reply.
  • Forgetting, sealing and the note ask, whenever they are needed.
  • Sleep after a long silence.
  • Scoring new facts and indexing what left the view.
meter 2 · per GB-month

Stored memory

Price to be announced

What your memories hold, measured once a day per project and averaged over the month. Deleting a memory stops it counting from the next measurement.

  • Every message and reply, as sent.
  • Facts written by the model or the system.
  • Kept for as long as the memory exists; nothing is aged out.
Not metered

What costs nothing.

The meters count work that grows with your conversations. These do not.

  • decorate calls inside a tool loop: the reply and the tool results, with no new user message.
  • report, every time.
  • Creating, reading, exporting and deleting memories.
  • The fallback to the local tail when we cannot be reached: it runs in your process.
free tier

Start free

A monthly allowance of user turns and a number of memories, with no card. When an allowance is used up, decorate answers 429 quota_exceeded and the SDK falls back to the local tail, so your agent keeps answering. The allowances are set with the prices.

enterprise

Talk to us

Volume pricing, invoicing, a data processing agreement, and questions about where data lives. There is no SLA yet; tell us what you need.