Deploy an Agent as a Public Chatbot

Before You Start

Deployments are available on Pro and higher plans. You also need a finished agent that has been tested in the workspace before you share it with visitors.

If the chatbot should answer from Gmail, Slack, Linear, Stripe, docs, or another business system, connect those apps and confirm the agent can use the right tools before deployment.

Decide who the chatbot is for, what it is allowed to answer, which actions need approval, and whether visitors should be anonymous or signed in.

Open the Deploy Page

Open the agent you want to publish, then choose Deploy. The direct route is /agents/[id]/deploy, where [id] is the agent id.

Choose New deployment to create a public chatbot configuration for that agent. Use one deployment per audience or access model so settings stay easy to review later.

Choose the Slug and Public URL

Enter a short, readable slug for the chatbot. The slug becomes the public URL, such as your-slug.agenticworkers.com.

Use a slug that visitors will recognize and that you are comfortable sharing publicly. If the slug is already taken, choose a more specific version for your workspace, brand, team, or use case.

Set Access and Privacy

Choose anonymous access when anyone with the link should be able to start chatting without signing in. Choose sign-in required when each visitor should authenticate before using the chatbot.

Set privacy to public for an openly shareable chatbot. Use password protection when visitors should enter a shared password, or use an email allowlist when only specific addresses should be able to access it.

Match the privacy setting to the data and actions available to the agent. Customer-facing and internal workflows usually need stricter settings than a simple public FAQ bot.

Choose Payment and Memory Options

Choose who pays for usage. Owner pays means visitor chats consume the deployment owner workspace credits. Visitor pays means visitors use their own account or payment path when required by the deployment.

If your plan supports it, enable per-visitor memory when the chatbot should remember context separately for each signed-in visitor. This option is available on Ultra and higher plans.

Leave per-visitor memory off when the agent should treat every visitor session as fresh, or when you have not yet reviewed the privacy implications of retaining visitor-specific context.

Save and Share

Review the slug, access mode, privacy setting, payment model, and memory setting, then save the deployment.

After saving, copy the public URL and share it with the intended audience. For protected deployments, share the password or access instructions through a separate trusted channel.

Verify the Chatbot

Open the public URL in a fresh browser session. Confirm the page loads at your slug, the access or privacy gate behaves as expected, and a visitor can start a chat.

Send a realistic visitor question and confirm the agent answers with the right instructions, tools, and boundaries. If the response depends on connected apps, review the trace, the step-by-step run log, or app result before wider sharing.

Troubleshooting

If you cannot deploy the agent, confirm the workspace is on a Pro or higher plan and that you have permission to manage the agent.

If per-visitor memory is greyed out, upgrade to Ultra or higher, or deploy without per-visitor memory.

If visitors cannot open the URL, check the slug, active deployment status, privacy mode, password, and email allowlist.

If the chatbot cannot use an app or tool, reconnect the integration, enable the tool for the agent, and test the agent inside the workspace before saving or sharing the deployment again.

If usage is higher than expected, switch to a stricter access mode, review who pays for visitor chats, or pause sharing while you check plans and limits.

Related Guides

Read "Manage Credits and Plan Limits" before sharing a high-traffic chatbot so you understand plan requirements, credits, and usage controls.

Read "Connect Apps and Tools to Your Agents" when the deployed chatbot needs access to Gmail, Slack, Linear, Stripe, or another integration.

Read "Give Your Agents Long-Term Memory" before enabling memory behavior that affects future conversations.

Read "Read an Agent Trace" when a visitor chat behaves differently from your workspace tests.