Chat Features
File attachments
You can attach documents to any message. Architect extracts the text and uses it as context for that turn — for example, to ground test generation in a spec you upload, or to inform a plan based on requirements.
Supported formats:
| Category | Extensions |
|---|---|
| Documents | .pdf, .docx, .pptx, .xlsx |
| Plain text | .txt, .md, .csv |
| Data | .json, .yaml, .yml, .xml |
| Web | .html, .htm |
| Code | .py, .js, .ts |
Limit: 5 MB per file. Multiple files per message are supported.
To attach a file, click the paperclip icon in the input bar.
Attachment content only applies to the message it’s attached to. If you want to reference the same document in a later message, attach it again.
Auto-approve
By default, Architect asks for confirmation before each action that modifies the platform (creating an entity, running a test set). The Auto-approve toggle, visible in the chat header, skips these confirmations until you turn it off.
Use auto-approve when you trust the plan and want to move quickly. Architect still reports each step as it happens — you just don’t need to confirm them individually.
The Accept and Change buttons appear only when Architect actually tried to call a confirmable tool in the current turn and the backend blocked it for your approval. They do not appear on informational replies or auto-resumed task summaries.
The auto-approve toggle does not reset automatically when you start a new conversation — switch it off when you no longer want actions to run unconfirmed.
Plan panel
The plan panel appears between the chat and the input when Architect proposes a plan. It shows the test suite as a collapsible checklist with a progress bar.
As Architect creates each entity, the corresponding checkbox is ticked. The panel gives a live view of creation progress without scrolling through tool activity in the chat. Architect also tracks whether prerequisites are ready or blocked before it generates test sets, so the checklist reflects behaviors, metrics, and behavior-metric mappings before generation starts. When everything is done, the panel shows a Plan Complete badge.
Live task progress
Long-running background work, such as endpoint exploration, appears inside the same assistant message bubble as regular streaming tool activity. While a task is running, Architect shows progress rows such as the strategy being run or each probing turn. When the background task finishes, the progress trail collapses and a Done. marker remains.
This keeps the chat anchored to the message that started the task instead of creating a separate progress bubble.
Mode chip
The mode chip in the chat header shows the current phase:
| Chip | Phase |
|---|---|
discovery | Architect is exploring your endpoint |
planning | Architect is formulating or refining a plan |
creating | Architect is building entities on the platform |
executing | Architect is running tests and waiting for results |
→ See Workflow for what happens in each phase.
Session sidebar
Past sessions appear in the left column of the Architect page. Each session is named automatically based on the initial request. Click any session to restore it and continue the conversation.
Sessions are scoped to the organization and the active project. Members of the same organization see the same session list within a project.
Suggested prompts
On a new session, the welcome screen shows a set of suggested prompts to help you get started. Click any chip to send it as your first message.
These suggestions cover the most common entry points: exploring a new endpoint, analyzing a past run, or kicking off a quick test.