ai
Agents
Use the agent to analyze crawls, attach context, run tools, and create artifacts from your data.
Where to find it
Open the agent chat with ⌘I from any workspace. Each workspace has its own thread list in the sidebar. Expand to fullscreen Agent view for chat plus the artifact canvas side by side.
Start a new thread with ⌘⇧N when a workspace is active. The start-screen helper answers crawl setup questions before you run a crawl.
What the agent can use
The agent reads your active workspace crawl — rendered content, metadata, filters, issues, link graph, semantic clusters, and embeddings. It does not guess from pasted URLs alone when crawl context is available.
Attach context to narrow the conversation before you send a message.
| Attachment | How | Use for |
|---|---|---|
| Page | Drag a table row, @ mention, or attach from search | Single-URL analysis |
| Multiple pages | Multi-select + attach | Compare or pattern-finding |
| Issue | From issue menu or inspector | Fix planning for one rule |
| Cluster | From semantic map | Topic-level summaries |
| Compare workspace | Composer compare menu | Cross-site analysis (paid) |
| Document excerpt | Select text in an artifact | Rewrite or action items |
What to ask
Ask for outcomes, not tool names.
| Goal | Example prompt |
|---|---|
| Prioritize work | "What should I fix first on this crawl?" |
| Issues | "Summarize critical issues by theme" |
| Content | "Which pages in this cluster are thin?" |
| Retrieval | "Which pages best answer [query]?" |
| Deliverables | "Create a sheet of pages missing meta descriptions" |
Starter prompts appear in an empty chat. Attach context to get prompts tailored to that page, issue, or cluster.
Tools the agent uses
You do not call these directly — the agent picks them based on your question.
| Category | Examples |
|---|---|
| Research | Query pages, semantic search, get issues, link graph, clusters, page evidence |
| Filters | Add or remove table filters (`manage_filters`) |
| Artifacts | Create or update sheets, documents, and charts |
| Schema | Read crawl field definitions, run SQL views on crawl data |
Destructive or high-impact actions may ask you to confirm before running.
Artifacts
The agent can create workspace artifacts during a thread:
| Type | Purpose |
|---|---|
| Sheet | Filtered URL lists, audits, tabular exports |
| Document | Narratives, briefs, fix plans |
| Chart | Visual summaries from crawl metrics |
New artifacts open in the canvas (fullscreen agent or inline workbench tab). Reopen saved artifacts from Artifacts in the sidebar or from artifact tabs in the title bar.
Export from the artifact canvas or Artifacts library: sheets and charts as CSV, documents as Markdown.
Filters and the table
Ask in plain language — for example, "show pages with missing titles" — and the agent adds table filters. For precise control, use the command bar yourself or combine both: filter manually, then ask the agent to analyze the visible set.
Compare workspace
Attach a competitor workspace in the composer to compare crawls in one conversation. Requires a paid plan. Useful for gap analysis and side-by-side topic coverage.
Search playground guide → (compare also works in semantic search)
Threads and versions
Threads are scoped to the workspace. Switch crawl version from the title bar if you need the agent on an older run — the agent uses whichever version is active in the workspace.
Tips
- Attach pages or issues before asking — less ambiguity, better answers.
- Use sheets when you want a reusable, sortable output.
- Use documents when you want prose for a client or team.
- Open Issue registry or the issue reference when you need rule names; ask the agent to group by theme or severity.