Managed Agents adds Amp, arm64 sandboxes, custom sandbox images, and a read-only Depot token

Updates and new features for Managed Agents:

• Managed Agents now runs Amp threads. Create an orb provider in Amp's Settings > Orbs, then paste its token into Settings > Connections > Agent providers in Depot. Each orb runs the agent's tools in a Depot sandbox with the connection's run settings, and paused orbs keep their files. See the Managed Agents - Amp docs for setup.

• Runs can now use arm64 sandboxes. Choose amd64 or arm64 in each provider connection's run settings under Settings > Managed Runs. This works with every provider.

• Runs can now start from your own sandbox image. To start with your toolchains and dependencies already installed, build the image as a Depot snapshot in your organization's registry. Then choose it in the connection's General section under Settings > Managed Runs.

  • Claude Code connections can also use a different sandbox image and setup script for each repository. Runs on any other repository use the connection's image and script.

• You can now turn off Depot Cache per provider connection. Depot Cache is still on by default. Turn it off for a connection with the Depot Cache switch in the connection's run settings. Runs on that connection then start without cache credentials.

• OpenAI and Devin runs can now use your organization's MCP servers. Devin needs a one-time step in Devin's settings first. See the Managed Agents - Devin docs for details.

• The Depot CLI is now signed in as your organization in runs from every provider. To give runs more access, such as starting container builds, add your own token as a run secret named DEPOT_TOKEN. To give runs less access, set the connection's Depot API token role to Read-only. The CLI can then read CI runs, jobs, logs, and test results, but can't start, cancel, or rerun CI runs.

See Managed Agents for setup and limits.