Skip to main content
On a personal Avibe paired with avibe.bot, Memory works out of the box with Avibe Cloud selected. You do not need to configure a model or API key.
When you turn Memory on, your Workbench messages and authorized users’ private messages are sent to Avibe’s cloud models to build memories. Memory searches send their queries to the cloud service too. If IM attachment processing is available, eligible attachment copies are also sent for processing. The local Memory store remains on your machine, but the cloud service handles the content sent to it under its policies.

​
Use the personal default

  1. Open Settings > Memory.
  2. If prompted, install or repair Memory runtime from Settings > Dependencies.
  3. Leave Avibe Cloud selected and turn on Memory.
Avibe uses the personal free pool. If the required cloud models are temporarily unavailable, processing pauses while new messages remain queued locally, then resumes automatically when the models return.

​
Use custom endpoints

Personal installations can use their own providers instead:
  1. Open Settings > Memory.
  2. Select Use custom endpoints.
  3. Configure both required endpoints:
    • Language model: an OpenAI-compatible endpoint address, model, and API key.
    • Embedding: an OpenAI-compatible endpoint address, model, and API key.
  4. Optionally configure Reranking and IM attachment processing.
  5. Save the settings. If the Embedding model or endpoint address changed, confirm the local index rebuild.
Custom endpoints connect directly to the providers you choose. They handle Memory inputs and all content derived from them under their own policies. This currently includes messages, search queries, distilled Memory snippets or candidates sent for reranking, and eligible attachment copies. Select Use Avibe Cloud to return to the personal free pool. Review the complete network boundary inventory before choosing a mode.

​
Upgrading an existing installation

Upgrading never replaces a complete manual Memory configuration. An installation that already has the required custom endpoints stays on Custom endpoints. Avibe Cloud becomes the default only when no complete manual configuration exists.

​
When your organization manages models

An enterprise-managed installation shows Models are managed by your organization instead of the custom endpoint editor. Your organization chooses the models, but these controls remain local:
  • the Memory on/off toggle;
  • Restart engine; and
  • Clear Memory Data.
When you turn Memory on, the Memory content described above is sent through avibe.bot to the providers configured by your organization. See the network boundary inventory for the complete enterprise boundary. When an organization first supplies both the Chat and Embedding models, Avibe pauses processing and asks you to confirm a local index rebuild. Messages already waiting to process remain queued. If the organization has not supplied both models, an existing working custom configuration continues until an organization-managed replacement is ready.

​
What stays local

Your distilled memories and search index stay on this machine. The local queue also keeps messages and eligible attachment copies while they wait to process. Clear Memory Data deletes Avibe-managed Memory data on this installation, including profiles, memories, indexes, and messages waiting to process. It does not delete original chat histories and cannot take back content already sent to a cloud model or custom endpoint. See Model Service for how personal and organization-managed models are supplied.