Canvases

A canvas is an interactive app that an agent builds around your PostHog data. A canvas can be disposable (one time use) or durable (something you come back to). Use one when a question or a workflow needs more than a static dashboard.

You describe the outcome. The agent writes the interface and the queries, then publishes the canvas to a space. The canvas runs inside PostHog against live project data, so your team opens the same working app instead of a screenshot or an export.

Canvases can combine saved insights with new queries, add filters and date controls, remember shared state, capture events, and offer approved actions.

When to use a canvas

Dashboards and notebooks in PostHog Web still do their jobs, and the PostHog MCP is still the fastest way to pull data into an agent chat. A canvas is for when the view needs its own layout, controls, or workflow, when the answer is easier to see than to read, or when a team needs to act on it together.

Ask the agent for a canvas when you want to do things like:

• Build a doc, deck, or checklist that pulls in live data
• Add filters, controls, or logic an insight can't express
• Combine context from multiple tools and external sources into one view
• Pair data with an action, like letting an owner mark a problem as reviewed

Limitations

A canvas is a sandboxed browser app, not a hosted service. These limits apply to every canvas.

Build a canvas

Open a space in PostHog Desktop, start a new session, then describe what you need. You can also ask for a canvas in any active agent session, or from a self-driving report.

A useful first prompt includes:

• The question or the workflow the canvas supports
• Who uses it
• The data or the saved insights it shows
• The filters and the controls people need
• Any action a viewer can take

Build a canvas that shows weekly signups, activation rate, and the top five
referrers for the last 90 days. Use our saved insights where they exist, and
add a date picker at the top. Add a breakdown by plan below the activation chart, 
and compare each number with the previous period.

What you can build

A canvas can be a data board, a document, a deck, a form, or a small tool. The patterns below work well. Copy a prompt, then change the details for your product.

Release health

Show how a release performs across adoption, quality, and support.

Build a release health canvas for our latest version. Show adoption, new
errors, support volume, and open pull requests. Add a version filter, and link
each problem to its source.

Customer lookup

Give support and success teams one place to understand an account.

Build a customer lookup canvas. Let me search by email or company, then show
recent activity, plan, feature access, errors, support conversations, and
relevant session recordings.

Weekly product review

Turn a repeated reporting task into a shared view with current data.

Build our weekly product review. Show activation, retention, feature adoption,
experiments, and the largest changes since last week. Add a date picker and a
short summary for each section.

Feature rollout

Connect rollout progress with behavior, quality, and feedback.

Build a rollout canvas for this feature flag. Compare exposed and unexposed
users, show conversion and errors, include survey feedback, and link to the
experiment and the flag.

Team workflow

Combine data with actions when a team must act from one view.

Build a triage canvas for our product team. Group current problems by
priority, show the evidence behind each one, and let an owner mark an item as
reviewed.

Document or deck

Write prose, a newsletter, or a slide deck that keeps its numbers current.

Build a monthly business review as a deck canvas. One slide for revenue, one
for growth by product, and one for churn risk. Pull every number from our
saved insights, and add speaker notes under each slide.

Improve a canvas

Select Edit on a canvas to open the chat panel and prompt changes from the original session. You can also edit a canvas from any other agent chat. Refer to the canvas by name or link. Teammates in a shared space can comment on part of a canvas. Double click an element on the canvas, then select Add comment. The comment stays in the canvas thread, where you can use it to prompt the agent.

Each publish creates a version. Open an earlier version from the version list, and select Revert to this version when a revision does not work.

Share a canvas

A canvas belongs to a space, so teammates find it beside the sessions and the context that produced it. Everyone with access to the space and its PostHog project can open the canvas, comment on it, and ask for changes.

Select Copy link to get an HTTPS link. The link opens the canvas in PostHog Desktop. A canvas in your personal space stays private, so build it in a shared space when the team must use it.

Data and actions

Canvases use the PostHog project selected in Desktop. They can query the same project data available to the agent, including saved insights and supported PostHog products.

Each canvas declares the data and the actions that it needs, and PostHog limits the canvas to that list. Review the request when an agent builds or updates a canvas, especially when the canvas can capture events or change data.

See PostHog integration for project selection and data access. See Spaces for shared context and collaboration.