Overview
Some organizations use multiple LLM models that are fine-tuned for different tasks. Currently, the initial AI Copilot release supports multiple models, but they cannot be used for the same function across the workspace and end users cannot opt to change the default model for AI Copilot interactions once configured in the system console.
A differentiator with Mattermost’s AI plugin is the flexibility being offered. The aim is to enhance that flexibility even further by allowing multiple LLM models in the following ways:
Configure multiple bots/models for use throughout the workspace
Provide the ability for end users to see the bot their interacting with and change that if there are multiple bots configured in the workspace
Provide ability to tag/mention different bots to engage with specific pre-configured LLMs
Configuring bots/models in the system console
...
Collapsed - hover state
...
Expanded state
...
...
For each bot added to the plugin, admins can configure the following:
Display name: How the bot will display in the configuration and how users will see the bot displayed
NOTE: This does not change the “AI copilot” name - this is just the name that is displayed in the bot tag/dropdown
Service: The LLM provider (e.g. Open AI, Anthropic, etc)
API Key: The API key provided by the LLM provider
Default Model: A model that you can access with your API key (e.g. gpt-4)
Bot icon: The bot’s avatar image that will be displayed for replies directly from the bot
Team access: By default all teams are added, but admins can remove specific teams to restrict access to specific bots
Instructions: Admins can provide specific instructions to fine-tune the bot responses in order for the bot to respond or act in specific ways.
Enterprise-only notice
For non-enterprise workspaces, they cannot add multiple bots and a notice will display.
...
If there is only one bot configured in the system console, we don’t display the dropdown/label to end users - except in the user profile popover.
...