Changelog
You can now run automated test conversations against your AI agent to make sure it behaves as expected.
Run tests anytime, especially after updating your agent’s instructions, model, data sources, procedures, or actions, to catch issues before they reach customers.
For actions, you can:
- Check expected actions: Optionally verify that the agent called exactly the actions you listed.
- Mock responses: Return your own data or an error to see how the agent responds to different scenarios.
- Run live actions: Call supported read-only actions, such as Shopify product lookups or web search. Write actions and custom actions can’t run live.
Every paid plan includes free monthly test credits: 200 on Hobby, 500 on Standard, and 1,000 on Pro and Enterprise. On the Free plan, tests use message credits.
Backstage is your agent's AI-powered operations center. Manage your agent through conversation, right from your Chatbase dashboard, Slack, or iMessage.
What you can do
- Monitor performance: Analytics, conversation trends, credit usage, and knowledge gaps.
- Manage knowledge and actions: Add, update, or remove sources, actions, and integrations.
- Build custom actions: Describe what your agent should do, and Backstage builds and tests it.
- Configure your agent: Instructions, model, temperature, and appearance.
- Manage channels: Enable or disable where your agent is live.
- Review your Helpdesk: Tickets and performance metrics.
- Audit your agent: Scheduled or on-demand reports with issues and suggested fixes.
Backstage always shows you what it plans to change, and nothing is applied until you approve it.
Custom actions
Describe the action and share your API docs if you have them:
"I want my AI agent to check a flight's status using the booking reference. Here are our API docs: [link]"
Backstage drafts the action, then asks for your approval to test it and turn it on.
- API keys stay private: Keys are entered in a masked field and sent directly to the action. Backstage never sees them.
- Safe testing: Nothing is sent until you click Run. You see the exact request first, with warnings for methods that change data.
Agent audits
Ask Backstage for a daily, weekday, or weekly audit:
"Send me a daily audit at 9 AM. Lead with refund requests, and flag anyone asking about shipping to Canada."
Each audit reviews all activity in the reporting window, including conversation quality, knowledge gaps, instructions, training status, procedures, actions, and your Helpdesk queue. Reports highlight what needs your attention, what's working well, what changed since the last report, and suggested fixes.
Reports arrive in Slack or iMessage if connected, or in Backstage with an email notification. Reply to dig deeper or ask for a fix. To run one anytime, just ask Backstage to "audit my agent."
We’ve added Claude Opus 5.5 and Claude Sonnet 5.5 to Chatbase, giving you access to two of Anthropic’s latest models.
You can now choose between Claude Opus 5.5 for more complex tasks and Claude Sonnet 5.5 for a balance of speed and performance, giving you more flexibility to pick the right model for your AI agent.
We’ve added GPT-6.1 Sol to Chatbase, giving you access to another powerful model for your AI agents. With stronger reasoning for more complex tasks, GPT-6.1 Sol costs 2 message credits per response.
Chatbase Procedures let you teach your AI agent how to handle repeatable workflows, much like writing an SOP for a teammate.
Write the process in plain language, including when it should apply, what information the agent should look up, and which actions it should take. Your agent can then follow the same process consistently across conversations.
Use Procedures for workflows like qualifying leads, handling returns and refunds, troubleshooting issues, or any process where your agent needs to follow a specific set of steps.