The ActVox agent works from workspace context, attached material, relevant pages, and supported integrations. It is designed to create durable results - pages, plans, drafts, records, and artifacts - rather than only returning disposable chat prose.
Ask good questions
Use a clear outcome, relevant context, and a boundary. For example:
Review the competitor notes in this workspace, create a one-page positioning recommendation, and do not change our existing strategy page.
You can reference pages, files, and workspace objects in a conversation. Keep source material current so the agent can ground its output.
Attach files to a conversation
Use the attachment control in the composer to add authorized source material to the current request. ActVox uploads the file to durable workspace storage before it sends the message. While an upload is in progress, Send and the Enter shortcut remain disabled. Wait until the attachment appears in the composer; if the upload fails, resolve the visible error and attach the file again.
The agent can extract bounded text from PDF, DOCX, TXT, Markdown, CSV, JSON, XML, HTML, and other text-based files. Images, audio, and supported video use their media-processing paths. A successfully attached file keeps a persistent serving URL in the conversation, so the message and attachment remain available after reload. Extraction still depends on a readable file: scanned PDFs without text, encrypted documents, corrupt files, and unsupported binary formats may require another format or explicit context in the prompt.
Tell the agent what the file is and what to do with it. For example:
Read the attached customer interview notes, identify recurring objections, and create a cited summary page. Do not infer facts that are not present in the document.
Do not attach credentials, secrets, or material you are not authorized to process.
Plan
Use Plan for work that is multi-step, ambiguous, cross-functional, or consequential.
A plan should make the proposed work inspectable before execution. It can include scope, evidence to use, assumptions, steps, expected outputs, risks, and approval points. Plan is appropriate for requests such as:
- Create a 90-day content strategy from research and the current brand profile.
- Audit competitor positioning and recommend a category narrative.
- Design a research and reporting workflow before changing a database.
Review the plan, correct assumptions, and approve the next step in the conversation. Do not treat a plan as proof that an action has been completed.
Act
Use Act when you want the agent to execute a supported operation. Act can create or update supported workspace objects, organize material, produce reports, generate drafts, and run eligible analysis flows.
ActVox verifies required effects before it claims success. For example, when asked to create a page or content draft, the expected result is a real workspace object you can open and edit - not only a pasted response in the conversation.
Low-risk, reversible workspace writes may run directly when your request is explicit. Destructive, bulk, publishing, live-document, or otherwise high-impact actions can require review or approval. External availability also depends on integration permissions and product policy.
Skills
Skills are reusable, versioned instruction packages available in a workspace. Access them from the chat composer’s ✨ menu or through their associated slash commands when available.
Use a skill when the task benefits from a repeatable procedure, such as onboarding context capture, creating a company profile, or producing an initial content draft.
Good practice:
- Read the skill description and expected output.
- Confirm it matches the workspace and intended outcome.
- Run it with the specific information it needs.
- Review the generated page, draft, or artifact before sharing or publishing.
Skills do not bypass workspace permissions, action allowlists, or approval requirements. User-authored skills are versioned and should not be silently rewritten by autonomous improvement workflows.