How it works

When you submit a request to Storytell, the agentic AI system initiates a 4-step process that operates independently to fulfill your needs. The system first analyzes your request to understand intent and complexity, formulates a strategic approach, executes that strategy using appropriate tools, and generates comprehensive artifacts that can be automated for future use.

Each interaction with Storytell is unique because the AI agent autonomously decides what steps to take based on your specific request, rather than following a predetermined template.

Step 1: Strategy Planning

Storytell analyzes your request for intent and complexity and plans a comprehensive strategy to fulfill your needs. During this phase, Storytell performs detailed internal reasoning, including complexity assessment, goal analysis, and strategic approach formulation.

What Storytell is doing:

• Converting broad business questions into structured analytical frameworks with clear reasoning

• Determining optimal research and analysis approaches for complex questions

• Identifying required data sources and analytical methods

• Displaying "Tools I'm choosing for this job:" with detailed descriptions of each selected tool

Step 2: Tool Execution

Storytell executes its planned strategy by calling multiple tools in iterative rounds. This may include sequential knowledge base searches, web searches, and other specialized tools, with real-time status updates showing search progress and results.

How LLMs behave during this phase:

• If you ask about something in your uploaded assets, the LLM will stay within your knowledge base

• If you explicitly request new or “latest” information, it will trigger a web search

• For complex prompts, the LLM may combine both, grounding in your assets first and then augmenting with external results

What Storytell is doing:

• Conducting comprehensive research with multiple search iterations for thoroughness

• Analyzing both internal knowledge base data and external web sources

• Tracking search progress and understanding which sources contribute to your output

• Displaying detailed search results with specific counts (e.g., "Found 24 knowledge base results")

• Showing completion times for each search operation

Step 3: Evaluation

This is a critical step in which Storytell reviews the completeness and quality of the gathered information. Storytell determines whether sufficient data has been collected or if additional searches are needed before proceeding.

What Storytell is doing:

• Conducting comprehensive information validation before beginning analysis

• Performing self-assessment of data completeness and identifying potential gaps

• Implementing built-in quality control mechanisms to validate information sufficiency

• Reviewing the execution plan to ensure all necessary research steps have been completed

Step 4: Artifact Generation and Results

Storytell creates an editable artifact for your results.

Understanding LLM differences in Agentic Mode

Different LLMs handle this process in slightly different ways:

• Sonnet: Sequential reasoning, searches knowledge base first, then expands to the web with multiple refinement steps

• Gemini: Faster, often parallel execution, may mix knowledge base retrieval and web calls simultaneously

• GPT: Offers control (e.g., “limit searches to two in parallel”) and supports stoppable runs

• All LLMs: Use the current date to ensure web results are time-relevant

This ensures your experience adapts to the intent of your prompt, giving you the right balance of grounded knowledge and fresh external context.

Understanding the process

Real-time transparency

Throughout the process, you'll see real-time status updates including "Planning...", "Digesting...", "Evaluating...", "Analyzing...", and "Writing..." that show exactly what Storytell is doing at each moment.

Credit tracking

Storytell displays credit usage for each major phase (e.g., "28.10 credits", "7.42 credits"), allowing you to understand the computational cost of different analysis types.

Iterative intelligence

StorytellI may perform multiple rounds of searches and evaluations, continuously refining its approach based on what it discovers, rather than following a linear path.