Skip to main content
Agent Builder v2 is in open beta. Features and limitations described here are subject to change.
Agent Builder v2 is Writer’s full-code agent deployment experience. You build agents in your own IDE with the frameworks you already use, then deploy them to the Writer platform with built-in supervision and governance. This page explains what Agent Builder v2 provides, how it works, and the core concepts you need before you deploy your first agent. With Agent Builder v2, you can:
  • Deploy custom agents from your IDE or AI Studio with no framework lock-in.
  • Keep separate development and production environments, each with version history.
  • Manage environment variables for API keys and secrets.
  • Use MCP server integration to deploy and manage agents directly from your IDE.
  • Add Writer’s observability and agent supervision on top of your existing code.

​
Understand how it works

You bring your application code with a Dockerfile. Writer handles the containerized hosting, environment management, and deployment pipeline, so you keep full control of your code and development workflow. A typical flow looks like this:
  1. Create an agent in AI Studio, which generates a unique agent ID and a development environment.
  2. Connect your IDE to the Writer MCP server.
  3. Package your application and deploy a snapshot to the development environment.
  4. Test the development deployment, then promote it to production.
For the full walkthrough, see Get started with Agent Builder v2.

​
Supported frameworks

Agent Builder v2 deploys any dockerized Python or JavaScript/TypeScript application. The following frameworks are tested and verified:
If your framework isn’t listed, it can still work as long as it runs in a Docker container and listens on a port other than 8001 (which is reserved).

​
Key concepts

Understanding environments, snapshots, and the promotion flow helps you reason about how your code moves from development to production.

​
Environments

Each agent has separate development and production environments, and you can create additional custom environments. Each environment has its own:
  • Deployment URL, which you can change
  • Environment variables
  • Version history
  • Access controls (production only)

​
Snapshots and versions

Every time you upload code, Agent Builder v2 creates a new snapshot (version). For each agent you can:
  • View the full version history.
  • Roll back to any previous version.
  • Export a snapshot as a .tar.gz file.
  • Delete old versions.

​
Promotion flow

Promotion copies a snapshot from development to production. Environment variable keys are copied, but values must be set separately for production to maintain security boundaries. After promotion, the agent appears in the Agents tab in Writer App for you and anyone you shared it with.

​
Known limitations

Agent Builder v2 has the following limitations during open beta:

​
Frequently asked questions

Yes. Your agent runs as a standard container, so you can use any SDK, API, or model provider, including Writer models, OpenAI, Anthropic, and AWS Bedrock. Set the relevant API keys as environment variables.
Your agent can make outbound network requests to any public endpoint. Connectivity to private databases, such as Snowflake or internal APIs, through private link is in development. Contact Writer to discuss your requirements.
If your application runs in a Docker container and listens on a port other than 8001, it works. The supported frameworks list represents what Writer has explicitly tested.
For agents accessible to Writer users (only you, specific teams, or all teams), Writer handles authentication through AI Studio access controls. For custom authentication for external users, implement it within your application code, since Agent Builder v2 gives you full control of the request and response lifecycle.
Agent Builder v2 is for developers who want full code control and work in their own IDE. No-code agents are for business users who build with natural language. They share the same backend infrastructure but serve distinct personas.

​
Next steps