Skip to main content
Background Models are specialized LLMs that work behind the scenes in Tess, executing specific functions that go beyond the main response in the chat. They allow more control, better performance, and cost optimization, since each function can use a different model, chosen according to the need.

​
What is it?

In addition to the main model that responds in the chat, Tess uses 8 auxiliary models responsible for tasks such as:
  • Extraction and organization of memories
  • Task routing between agents
  • Coordination of multi-agent executions
  • Intelligent search
  • Error explanation
These models operate transparently for the end user and can be configured by Owners or Administrators to balance quality vs. cost.

​
How to use it?

  1. Access the workspace settings
  2. Go to the Background LLMs section
    Captura De Tela 2026 06 09 Às 17 14 06
  3. View the available functions (Memory, Autopilot, Workflow)
  4. For each function:
    • Select the desired model
    • Adjust according to quality or cost needs
  5. Save the settings and you’re done — the models will start operating automatically in the background\

​
Deeper explanation

Background Models are organized into three main blocks:

Persistent Memory (3 models)

  1. Extraction Model
    Reads conversations and identifies relevant facts about the user. In this case, the model runs automatically after each conversation. More capable and advanced models extract richer and more accurate facts.
  2. Collective Digest Model
    Runs daily and synthesizes what the team learned during the day. Transforms dozens of individual facts into a readable summary for managers. Uses only facts authorized to be shared.
  3. Consolidation Model
    Keeps memory clean over time. Merges duplicate or fragmented facts into a single, more complete fact. Prevents noise accumulation in memory.

Autopilot (4 models)

  1. Automatic Routing Model
    When a task arrives without a defined destination, it decides which AI Employee is most suitable to execute it.
  2. Team Coordinator Model
    In multi-agent executions, it decides the sequence and distribution of work — who acts first, who reviews, who delivers.
  3. Automatic Search Model
    Before each response with active search, it analyzes the question and decides whether it is necessary to search the internet — and which terms to search for.
  4. Hiring Model (Omnibar/hire)
    When you use /hire to create an AI Employee, it creates the agent profile and generates detailed work instructions.

Workflow (1 model)


Error Summary Model
When a workflow fails, this model translates the technical error message into simple and understandable language for the user — explaining what happened and what can be done.
Best practices
  • Use stronger models only where there is real gain
  • Keep lightweight models for repetitive tasks (e.g.: consolidation)
  • Review settings as team usage evolves
  • Test changes in controlled scenarios before scaling
  • Combine with the use of Memory to improve consistency

​
Important notes

  • Background Models consume credits, even when operating invisibly
  • Each function runs at different times (e.g.: real-time, post-chat, daily)
  • The choice of model directly impacts:
    • Quality of automations
    • Total workspace cost
  • Available only in plans with support for Background LLMs
Background Models allow Tess to operate with multiple specialized intelligences at the same time. This increases response quality, improves automation, and provides fine control over cost — all without complexity for the end user.