Deep Analysis

Documentation Index

Fetch the complete documentation index at: https://honeydew.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

Deep Analysis answers investigative questions — “why did revenue drop?”, “which segments are underperforming?” — by running a multi-step agentic workflow across your data. It breaks the question into sub-questions, executes targeted queries at each step, and synthesizes the findings into a reasoned conclusion with supporting evidence at every step.

​

How It Works

When you submit a Deep Analysis question, the agent:

1. Plans — decomposes the question into a sequence of analytical steps
2. Investigates — runs queries at each step, following the data
3. Synthesizes — combines the results into a conclusion backed by evidence from each step

Each step queries governed entities’ attributes and metrics from the agent’s domain, so results are consistent with your semantic model.

​

Context

Each analysis session runs within the scope of an agent, which provides two things:

• Domain — the set of entities’ attributes and metrics the AI can query. Only data exposed through the domain is accessible during analysis.
• Context items — analytical skills, organizational knowledge, and historical traces loaded into the session. These shape how the AI approaches the question: which methodologies it applies, what business conventions it follows, and how it interprets past events in the data.

The same domain can power multiple agents with different context sets. See Agents and Context Management.

​

Analysis Steps

As the analysis runs, the UI surfaces short progress hints for each step. You can expand any step to inspect the intermediate query results and the reasoning behind it. In Slack and Teams, progress hints appear while the analysis runs. The final answer is posted when the analysis completes. Via the GraphQL API and MCP Server, the call returns when the analysis completes. The response includes the full reasoning chain as categorized text (plan, interpretation, final_conclusion) alongside intermediate data results.

​

Charts

The agent generates charts as part of its output. It selects the chart type, layout, and legend based on the question and the data — no configuration needed. In the Honeydew UI, charts render inline within the analysis. In Slack and Teams, charts are delivered as static images. Via programmatic interfaces, chart data is returned for your application to render. See the GraphQL API and MCP Server reference pages for response formats.

​

Follow-Up Questions

Deep Analysis sessions are conversational. Follow-up questions build on the prior analysis — calculations and data from earlier steps stay in context. In the Honeydew UI, Slack, and Teams, continue in the same thread. Context is preserved automatically. When using the MCP Server, pass the conversation_id from a previous response to continue a conversation. Via the GraphQL API, reuse the same chat_id from the original create_chat call.

​

Feedback

You can rate and comment on Deep Analysis responses directly from the UI, Slack, or Teams. Feedback is stored and visible to the team responsible for maintaining the agent’s context and semantic model, so they can identify gaps and improve analytical quality over time.

​

Available Interfaces