Model Routing
Routing lets different kinds of work use another LLM account or model while the main chat remains on its default.
What routing means
Assistela does not call an LLM only for ordinary Chat answers. It also uses models to summarize older conversation, review memory, run the Personal Assistant, plan tool use, work with code, and perform diagnostics. Routing lets you assign a different account or model to each purpose.
By default every purpose is set to Default, meaning it inherits the main AI account and model. This is recommended until you have a specific reason to separate something.
Available purposes
- Chat: ordinary answers and the main conversation.
- Summarization: compressing older history as context grows.
- Conversation search: finding older messages and sessions.
- Memory review: evaluating and updating durable memory.
- Personal Assistant: periodic proactive checks.
- Reflection: evaluating results and follow-up steps.
- Action planning: deciding which tools to use and in which order.
- Coding: programming-focused work.
- Diagnostics: model checks and operational analysis.
A model that writes good prose may still call tools poorly. For Personal Assistant, action planning, memory, coding, and diagnostics, use a model with reliable tool calling.
Configure a custom route
Prepare both accounts
First add and independently test every account under Settings → AI Accounts.
Open routing
Choose Assistela → Settings → Advanced → Model Routing.
Select a purpose
For the required row, change Default to Custom.
Choose the connection, account, and model
Select the connection type, AI account, and exact model. If the model is absent, use a custom model ID only when you know the provider’s exact identifier.
Test that exact feature
Test a Chat route with a message, a Personal Assistant route with Run Now, and a summary route with a longer test conversation or diagnostics.
Sensible strategies
- Simplicity: leave everything on Default. This creates the fewest failure points.
- Lower cost: use a fast inexpensive model for summaries and conversation search, and a stronger one for Chat and action planning.
- Specialization: use a coding model only on the Coding route.
- Separate Personal Assistant: use a dependable tool-calling model with its own cost budget.
Every custom route is another dependency. If that account runs out of credit, its API key expires, or the provider renames a model, only part of Assistela may fail. Monitor usage for every provider involved.
Return to the default
Set the affected purpose back to Default. It will again inherit the main AI account and model. If the failure disappears, the cause was the specific account, model, or its tool-use ability.