Your AI agents,
as infrastructure
as code

It is the official Terraform provider for ChatBotKit. It lets you declare your entire AI agent stack — bots, datasets, skillsets, abilities, files, spaces, secrets, policies, portals, and channel integrations — as code. You run terraform apply and the platform provisions everything for you. The provider is published on the Terraform Registry at chatbotkit/chatbotkit.

Most AI platforms only give you a dashboard. You click around to build an agent, and there is no reproducible record of what you did. With ChatBotKit the agent itself — its backstory, its tools, its memory, its credentials, and the channels it speaks on — is a declarative resource graph. You version it in git, review changes in pull requests, promote it across environments, and tear it down cleanly. The agent's brain and behavior live in the same workflow as the rest of your infrastructure.

29 resource types and 4 data sources. That covers bots, datasets, blueprints, skillsets and individual abilities, files and file content, spaces and space storage, secrets, policies, portals, and 17 channel integrations including Slack, Discord, Telegram, WhatsApp, Messenger, Instagram, Microsoft Teams, Google Chat, Email, Twilio, Notion, Sitemap, Support, Trigger, Widget, MCP Server, and Extract.

Use one master token plus the provider's run_as attribute, which sends the X-RunAs-UserId header — the same pattern as the AWS provider's assume_role. Each provider alias targets a customer's sub-account, and a shared module deploys the same agent into every sub-account in a single terraform apply. No per-customer tokens and no for_each gymnastics. Each tenant gets a fully isolated agent and resources.

Yes, and the two models compose cleanly. You declare the agent's structure and starting state in Terraform, then give it abilities that let it evolve — a self-improving agent rewrites its own backstory file, a workflow agent discovers and installs skillsets on demand, and an MCP-aware agent searches a registry and installs new tool servers as it needs them. Terraform owns the baseline; the agent owns its growth from there.

Through the chatbotkit_secret resource. Secrets can be shared (a single machine-to-machine service token used by the agent itself) or personal (each end user signs in once through your SSO, and the agent then acts on their behalf with their own permissions). Template-based secrets wire up OAuth for platforms like Google, Notion, and others without you ever handling raw tokens in state.

Yes. It is a standard Terraform provider, so it works with plan, apply, destroy, import, modules, variables, outputs, remote state, and workspaces. It drops straight into your CI/CD pipeline alongside your AWS, GCP, Cloudflare, or Kubernetes resources. Reviewing an agent change becomes a normal pull request with a normal plan.

Yes. Use terraform import to bring existing bots, datasets, skillsets, and other resources under Terraform management, then continue managing them as code. You do not have to start from scratch to adopt infrastructure as code.

Yes. The provider is open source and the repository ships with more than a dozen reference architectures — multi-tenant SaaS, AI employees, second-brain assistants, workflow orchestrators, dual-agent systems, MCP factories, and self-improving agents — that you can copy, apply, and adapt.

Add the chatbotkit/chatbotkit provider to your configuration, set the CHATBOTKIT_API_KEY environment variable, write a chatbotkit_bot resource, and run terraform init followed by terraform apply. Your first agent is live in minutes, fully reproducible and under version control.