AI Assistants

Assistants are a core AI component within Dystr Workspaces. They act as specialized agents capable of performing various tasks. Think of them as virtual robots that run within your project sandbox. They can be configured for different tasks, working in conjunction with computations and scheduled jobs. They can be run one-off (Chat) or run repeatedly in the background (Workers).

‍

Assistant Types

Dystr currently supports two main types of Assistants:

Chat Assistants: These assistants operate in an interactive, request-response mode. You provide a prompt (e.g., a question, request or instructions), and they provide a textual response. This is ideal for quick queries, exploratory analyses, and generating initial ideas. Chat Assistants use natural language prompts for input.
Worker Assistants: Worker Assistants are designed to automate recurring tasks and integrate with external tools. They respond to events or scheduled triggers, acting autonomously to execute pre-defined instructions. They can be set up to perform specific actions (e.g., process data, generate reports, etc.), automatically triggered by events or scheduled jobs.

Context Construction

‍

Chat Assistants

Can access and modify Workspace resources.
Maintain an ongoing conversation context.

Worker Assistants

Can access and modify Workspace resources.
Each run includes the configured system prompt, any trigger arguments, and an otherwise clear context for each execution.
Focus on specific repeatable task execution rather than conversation.

Execution Flow

Both Chat and Worker assistants follow the same basic execution loop:

Start
- Receive input (message for Chat, trigger for Worker).
- Load context (conversation history for Chat, system prompt + trigger arguments for Worker).

Process
- Model processes input with context.
- Executes commands as needed.
- Updates Workspace state.
- Continues until model signals completion.

Complete
- Records results.
- Prepares for next execution.

The key difference is that Chat assistants maintain conversation history between executions, while Workers start fresh with just their system prompt and trigger arguments for each run. This makes Chat ideal for exploratory interactions, and Workers great for consistent, repeatable tasks.

‍

Assistant Commands

Assistants can execute a set of commands to interact with the Workspace. See the Assistant Commands reference for more details.

‍