AI memory · custom integrations and companions
AI memory.Your control.
Give your AI product useful context across conversations. We build memory that can be inspected, traced to its sources and managed by its user, then connect it to your assistant or application.
Illustrated memory flow
Illustration: without saved memory, these three example conversations have no shared context.
Fig. 02.3 — Memory inside a companion · illustration
A companion starts as a description in its creator's own words: who it is, how it speaks, what it cares about, where its limits are.
Useful context, carried forward.
An assistant can use a preference, project detail or other useful fact again when it is relevant. Our memory engine selects what to keep, retrieves it for a later conversation, and keeps the source behind each reflection.
One preference, used again.
A fictional example of the flow an integration can provide.
- 01 · Save
- The user says, “I prefer short, bullet-point summaries.” With memory enabled, that statement can be kept with its source.
- 02 · Recall
- In a later conversation, the user asks for a project summary. The relevant preference is supplied as context for the answer.
- 03 · Control
- The user can inspect the saved statement, submit a correction, or ask the product to forget it.
Remember what matters.
A conversation history alone does not decide which details matter later. Your product needs a way to keep useful context, handle a changed or conflicting statement, and let the user see what the AI is working from.
The parts we have built.
- Selective memory
Built and tested - The companion path keeps selected statements in the user’s own words. A repeat reinforces an existing memory; a paraphrase or contradiction can be held for review.
- Recall and reflection
Built and tested - Retrieve relevant memories by meaning and words. Reflections connect insights to the statements they came from, so their basis can be inspected.
- User controls
Built and tested - Inspect and search saved memories, give feedback, submit a correction, forget a memory or export it. In the CreatorsHub companion flow, memory is optional and off until enabled.
- Memory API
Built and tested - A separate service exposes memory and consent operations for integration into other applications. Access rules and product behavior are configured and checked for each integration.
Fit the memory to your product.
We agree what your product may remember, how permission works, how it handles corrections, and which controls users need. Add a character and voice when the product calls for a companion.
From a service to your application.
A memory integration in your existing application, or a complete AI product built around it. We connect the service, build the controls and verify the agreed behavior in your product.
See where it fits.
The interactive figure is an illustration. CreatorsHub shows how memory fits into a companion product; its companion screens require sign-in.
What should your AI remember?
Clear about what is kept.
Memory is selective, and a response can still be wrong. Forgetting removes a memory from use; chat history, audit records and backups may have separate retention rules. We define those rules with you before delivery.