Hey everyone,

We have a LOT of changes in this release, so buckle up.

Tasks

The agent will now break down complex changes into tasks that maintain their own context. Tasks help the agent stay on track by maintaining their own state independent from the rest of your chat.

Runners

We added a sandbox runtime that lets the agent run commands in your cloud environment. You can connect your cloud environment using your cloud credentials, then ask the agent to deploy your application or apply your terraform. The agent will then kick off and debug whatever commands are needed. Check out our guide on connecting your cloud to get started.

Environments

Your cloud environments are automatically tracked whenever the agent makes a change to your cloud on your behalf. This helps you keep track of what operations were performed on your cloud for a given chat, and even lets you clean up all resources that were created.

Multi-Agent

We’ve introduced several new agent modes: `planner`, `runner`, and `debugger`. These specialized agents are fine-tuned to performing specific tasks. For example when you first start a chat the `planner` will generate a step-by-step plan that is needed to accomplish the ultimate goal. Then when you begin executing the plan each individual step is handed off to a runner. The runner has scoped context and only cares about their specific task. If a task fails we hand it to the debugger which specializes in how to debug issues with your cloud.

Bug Fixes and Improvements

• Better Terraform docs - we added further processing to all fetched Terraform docs to reduce their token usage and improve readability for the agent
• Tool serialization - we now persist all tool calls so they wont disappear when reloading your chat / accessing old chats
• Terraform resource listing - we’ve added the ability for the agent to fetch the latest resources available from a provider allowing it to also know the most up to date set
• Fixes small bugs around when the Terraform LSP is triggered which would cause the LSP to report non-existent errors.
• Docs refresh - we’ve refreshed our docs with some better examples and a getting started guide