Skip to main content
Honeycomb’s Canvas is an AI-powered, collaborative investigation workspace. Canvas combines conversational AI, Honeycomb’s query engine, and a multiplayer infinite canvas to help you analyze observability data, understand your system’s behavior, and debug application issues in Honeycomb. With Canvas you can:

How it works

Technical integration

Canvas integrates directly with Honeycomb’s query engine and has access to:
  • Team information
  • Environment and Dataset metadata
  • Field schemas and sample values
  • Query execution capabilities
  • Trace visualization tools
  • Boards
  • SLOs and Triggers
  • BubbleUp for identifying outliers
Canvas translates natural language requests into structured queries using Honeycomb’s API and presents results in an accessible format with appropriate visualizations.

Context you add

Beyond what Canvas can access on its own, you can give it more context to work with. Skills tell Canvas how to work, connectors bring in your other tools, and memory holds facts about your systems.

Skills

Skills give Canvas instructions and workflows to follow, such as the steps your team takes to investigate a particular kind of issue. You can keep a skill private or share it with your Team. Team owners can enable or disable Team skills and built-in skills for everyone on the Team. Canvas uses a skill when it’s relevant to an investigation, and you can also use one directly by typing /skill-name in a chat.

Connectors

Connectors let Canvas use services such as PagerDuty, incident.io, Notion, and Linear during an investigation. Each person connects their own account, and Canvas reaches the connected service as them.

Memory

Memory holds facts about your systems that Canvas carries into your investigations, such as which dataset answers questions about a payment queue. Canvas saves memories during an investigation, and you can view, add, edit, and delete them yourself. Your memories are personal. Memory is in beta.

Query transparency

Canvas makes all generated queries available throughout your investigation. You can examine the exact query syntax Canvas creates from your natural language questions and modify queries directly in the Query Builder.

Persistent workspaces

Canvas auto-saves investigations as persistent workspaces that you can revisit and share with Team members.

Your chat and team chat

Each Canvas investigation has Your Chat and Team chat tabs. Each chat is your own agent, letting you and your team have separate conversations with Canvas in the same investigation. Each chat also uses its owner’s memories, so Canvas brings your context into your chat and your teammates’ context into theirs. You can view team chats by selecting the Team tab then choosing a teammate from the dropdown menu.

Cursor chat

Cursor chat lets you send short messages that appear as bubbles near your cursor then fade away after a few seconds. It’s designed for quick, temporary communication during collaborative investigations. During an investigation with your team, press / to open the chat input and type your message (up to 64 characters), and press Enter to send.

Mermaid diagram support

You can create editable diagrams from Mermaid text in Canvas. Flowcharts, sequence diagrams, mindmaps, and state diagrams are supported.

Limitations

Canvas works within these boundaries:
  • Complex analytical questions may require iterative refinement.
  • Results are limited by the data retention policy of your Honeycomb account.
  • Canvas can’t modify your Honeycomb configuration or infrastructure.

Creating a Canvas investigation

Begin an investigation by creating a new Canvas or opening an existing query to use as context in Canvas. To start a new Canvas investigation:
  1. Select Canvas () from the navigation menu.
  2. Enter your question or investigation goal as a prompt.
  3. Submit () your prompt.
To start a Canvas investigation from an existing query:
  1. Build a query in the Query Builder.
  2. Select Canvas () in the Query Builder, then choose Open in Canvas.
Canvas creates a new investigation with your query as context.

Viewing past investigations

Return to an earlier investigation to review what Canvas found or to pick up where you left off.
  1. Select Canvas () from the navigation menu.
  2. Under All Investigations, find the investigation you want. To narrow the list, select the filter button next to All Investigations, then choose a filter. You can filter by:
    • Active investigations
    • Auto investigations
    • Trigger investigations
    • SLO investigations
    • Anomaly investigations
    • Team investigations
    • My investigations
Screenshot of the filter options for All Investigations in Canvas.

Adding Mermaid diagrams

Pasting Mermaid text into a Canvas creates a diagram you can edit. You can change the style of the entire diagram or specific nodes by selecting them and using the style editor. To resize a node or adjust the placement of a line, select it and use the handles that appear.
An example Mermaid diagram in Canvas. The whole diagram is selected and the diagram style editor is shown.
Here are some example diagrams you can copy and paste into Canvas. Try out different fonts, changing node shapes, or coloring different lines and nodes to make your diagram pop.

Example prompts

You can reference specific Datasets, Environments, fields, or include links to queries in your chat prompts. Use these examples as starting points to explore what’s possible in Canvas:
  • “What’s causing the spike in database request volume?”
  • “Investigate /link/to/a/trace/”
  • “What SLOs would you recommend for @my_dataset based on fields prefixed with my_prefix?”
  • “Which user cohorts are experiencing higher latency?”
  • “What fields are available in the frontend dataset?”
  • “Compare database query performance between production and staging.”

Effective question types

Framing your prompt as one of these question types helps Canvas understand what kind of answer you are looking for.
  • Diagnostic questions to identify causes of specific issues: Why did service X have high latency at 2pm yesterday?
  • Comparative questions to analyze differences between services or environments: How does the error rate in production compare to staging?
  • Trend analysis questions to examine patterns over time: Show me the pattern of database connections over the last week.
  • Correlation questions to explore relationships between metrics: Is there a relationship between cache miss rate and API latency?

Working with Canvas results

Canvas displays the queries it generates from your natural language questions in visualization panels alongside your chat session.

Interacting with visualizations

Interact with Canvas visualizations the same way you would with Query Results:
  • Select data points to view traces or examine the underlying query.
  • Hover over data points to explore details, such as timestamps and breakdowns of the selected dimension.

Editing generated queries

Each visualization panel includes options to examine the query syntax:
  • Select View query.
  • Select Open in new tab from the options menu ().
Both options open the query in Query Builder, where you can modify it to refine results.

Sharing a query

Share individual queries from a Canvas investigation by selecting the link icon () on any query panel. This copies a shareable link to that specific query.

Sharing a Canvas investigation

You can generate a shareable link to your Canvas investigation, so you can share it with your Team. To share a Canvas, select Share (). If the investigation is private, change its privacy setting first.

Best practices

Follow these guidelines to get the most accurate and relevant results from Canvas.
  • Be specific in your prompts:
    • Include specific service names when applicable.
    • Specify time ranges when looking at historical data.
    • Mention particular metrics or dimensions of interest.
    • State the relationship or pattern you are exploring.
  • Iterate on your investigation:
    • Start with a general question and refine based on initial results.
    • Ask follow-up questions to dig deeper into patterns Canvas identifies.
  • Provide context:
    • If switching topics, give Canvas enough context to understand the new direction.
    • Use mentions (@) to reference specific Datasets, Environments, or fields.
  • Ask for explanations:
    • Ask Canvas to explain query logic or data interpretation to deepen your understanding.