How Iris Works

Understanding the architecture behind autonomous AI agents

The Agent Loop

At the heart of Iris is an autonomous agent loop that transforms your natural language requests into concrete actions and results. Every interaction follows a structured cycle:

This loop can repeat multiple times within a single request. If the initial results need refinement, or if additional steps are required, Iris will continue iterating until the task is complete. This is what makes it truly autonomous -- it does not stop at the first answer.

Multi-Model Architecture

Iris does not rely on a single AI model. Instead, it uses a multi-model architecture powered by LiteLLM that routes each task to the most appropriate model:

• Claude -- Excels at nuanced reasoning, detailed analysis, and following complex instructions with precision.
• GPT-4 -- Strong at creative content generation, conversational tasks, and broad general knowledge.
• Specialized Models -- Vision models for image analysis (GPT-4V, Claude 3), image generation models for creating visuals, and code-focused models for technical tasks.

This flexibility means you always get the best possible output for each part of your workflow. The model selection happens transparently -- you simply describe your task and Iris handles the orchestration.

Tool System

Iris comes with 20+ integrated tools that give it the ability to interact with the real world. These tools are what transform Iris from a simple chatbot into an autonomous agent:

Information Gathering

• Web Search -- Query search engines and retrieve up-to-date information from across the internet.
• Browser Automation -- Navigate web pages, read content, and extract data from specific URLs.
• People Search -- Find professional profiles, company details, and contact information.

Content Creation

• Document Generation -- Create formatted PDF and DOCX files with professional layouts.
• Presentation Builder -- Generate slide decks with structured content and visual design.
• Website Builder -- Create web pages and deploy them from natural language descriptions.
• Image Generation -- Produce custom images, illustrations, and diagrams using AI models.

Data Processing

• Spreadsheet Analysis -- Process CSV files, generate charts, and perform data analysis.
• Vision Analysis -- Analyze images and extract information using multimodal AI models.
• File Processing -- Read and transform PDF, DOCX, and other document formats.

Context & Memory

Iris maintains full conversation context across every turn of a conversation. This means you can build on previous interactions without repeating yourself:

• Multi-Turn Conversations -- Reference earlier messages, refine outputs, and chain together complex workflows naturally.
• Tool Output Retention -- Results from previous tool calls are preserved in context, so Iris can reference research findings when creating a follow-up document.
• Conversation History -- Your past conversations are stored and accessible, allowing you to pick up where you left off.

Background Processing

Some tasks take longer than a typical request-response cycle. For these, Iris uses Dramatiq, a distributed task queue backed by Redis, to handle long-running operations in the background:

• Asynchronous Execution -- Complex research, large document generation, and multi-step workflows run as background jobs so the interface remains responsive.
• Real-Time Updates -- As background tasks progress, Iris streams status updates so you can track progress in real time.
• Reliability -- Background tasks are queued in Redis and processed by dedicated workers, ensuring that even if a step fails, the task can be retried gracefully.