Chatbase has expanded its developer API from agent interaction into agent administration. Teams can now create, configure, train, clone and delete AI agents programmatically, including the sources and presentation settings attached to them.
The change is most relevant to organisations running more than a handful of similar agents. Instead of reproducing configuration in the dashboard, they can make provisioning part of an internal tool, customer onboarding flow or deployment pipeline.
An agent can start from code
The new endpoints allow developers to create an agent and add training sources, including individual content sources or a website crawl. Training can then be triggered through the API and its status tracked until the updated knowledge is ready.
Agents can be listed and read, making it possible to build an inventory outside the Chatbase interface. A clone operation copies an agent with its sources, offering a base for templated deployments where most settings are shared but branding, instructions or knowledge differ by customer or business unit.
Chatbase also supports deletion of agents, sources and integrations. Those capabilities complete the lifecycle, but they deserve stronger safeguards than ordinary read operations. Production tooling should separate credentials, restrict deletion permissions and require an explicit confirmation or approval step.
Configuration reaches model and interface settings
Partial updates can change an agent’s model, instructions and temperature without replacing the full record. The API also covers widget configuration, allowed domains, rate limits, email collection and voice behaviour. Appearance settings can be managed alongside functional settings, which helps teams keep embedded agents consistent with a product or client brand.
This breadth turns configuration into something that can be reviewed and repeated. A team could store its approved defaults in code, apply them when a new agent is created and audit any differences. It also creates a risk of configuration drift if dashboard edits and automated updates are not reconciled, so organisations need a clear source of truth.
Freshness can be automated
The API includes a setting for automatic retraining every seven days. For agents grounded in frequently changing websites, that removes a manual maintenance task. It is still a fixed interval rather than proof that every important change has been incorporated immediately.
Teams should monitor training completion and retain the previous working state until a refresh succeeds. They should also test retrieval after substantial source changes; a successful training job does not guarantee that the agent will answer a particular question accurately or cite the intended material.
For more controlled publishing, a pipeline can trigger training after an approved content release, wait for completion and run a regression set before switching traffic to the updated agent. The expanded API supplies the building blocks, while the customer remains responsible for that quality gate.
Automation shifts the operational burden
Programmatic management can reduce repetitive dashboard work and make fleet-wide changes practical. It also means an incorrect script can affect many agents quickly. API credentials should be scoped and stored securely, calls should be logged, and bulk changes should begin with a small canary group.
The release moves Chatbase closer to an infrastructure component for organisations embedding purpose-built agents across sites or products. Its usefulness will depend not only on endpoint coverage, but on how teams wrap those endpoints with version control, testing and access policies appropriate to the scale of their deployment.