Get-started FAQ

Why can't I create another agent?

Creating requires the workspace's agent-create permission. The Create New Agent dialog also checks the active-agent limit. When the current plan has reached that limit, the name field is disabled and the dialog asks you to upgrade. You can instead reduce the active-agent count; an agent scheduled for deletion no longer counts toward the limit. Existing agents remain available from the Agents dashboard.

Where did my new agent go?

After creation, AgentStack opens the new agent's Analytics page. Return to the Agents dashboard to see every active agent. Selecting an agent card also opens Analytics; the card menu provides direct links to Settings and Analytics.

Which source should I add first?

Use the smallest authoritative source that answers a question you can verify. A single public policy page or focused Q&A pair is easier to evaluate than a broad website crawl. The Content page also supports files and Notion when those are the real source of truth.

Why doesn't the agent know content I just added?

Website ingestion is asynchronous. Under Sources → Content, wait for the website status to become Completed and inspect the Pages tab to confirm the relevant page was crawled and indexed. A source that is still Crawling or Processing is not ready for a final quality test. If the status is Failed, read the error shown with the source and correct that failure first.

Why did my Playground improvement disappear?

Apply & Test applies the current model and system prompt to the Playground test chat, but it does not persist the configuration. An Unsaved changes badge appears when the editor differs from the stored agent. Select Save to Agent after the candidate passes your tests. Reset restores the last saved values.

Why does the website widget behave differently from Playground?

Check four things:

1. The successful Playground model and prompt were saved to the agent.
2. Settings → General → Public is enabled; a non-public agent does not render its external widget.
3. Domain Protection allows the site's hostname, or is intentionally empty to allow all domains.
4. The site uses the current embed code from Deploy.

Then reproduce the question in a private browser window and look for the conversation in Activity. This separates cached page state from an agent-configuration problem.

Which model should I choose?

Begin with the default, currently GPT-5.2, and grade it against a fixed acceptance set. Switch when another model produces a repeatable improvement in factual accuracy, boundary behavior, or latency that justifies its credit cost. The selector is the current source of truth for available models, recommended labels, and credits. See Choose a model.

Does model comparison cost more?

Each active comparison panel sends a separate model request. Playground displays the combined credit cost and active panel count before a synchronized send. Pause panels you do not need, and compare on representative hard questions instead of spending credits on greetings.

Should product facts go in the system prompt?

Usually not. Keep stable behavioral instructions—role, tone, boundaries, uncertainty, and escalation—in the system prompt. Put policies, prices, procedures, and other changing facts in sources. This makes factual updates easier to own and reduces conflicts between instructions and knowledge.

What should an agent do when the answer is missing?

Define the expected behavior explicitly. A safe first-agent prompt says that the agent should acknowledge uncertainty and provide a configured support contact or manual next step when the available knowledge does not contain the answer. An actual transfer requires a handoff capability to be configured. Test this with at least one deliberately unsupported question before deployment.

Do I need to configure actions before launching?

No. A first agent can answer from sources without taking external actions. Stabilize answer quality first. Add integrations and actions later, with narrow permissions and deliberate testing, because actions can change another system rather than merely produce text.

What should I test before publishing?

At minimum, test several known facts, a paraphrase, an ambiguous request, an unsupported question, and a boundary case. Save the final configuration, then repeat a supported and unsupported question on the real customer channel. The response-quality guide provides a reusable checklist.

Where should I go next?

Follow Create your first agent for the complete setup sequence. Then use Sources for deeper knowledge configuration, Web widget for installation, Channels for other customer surfaces, and Inbox for human handoff.