Manifold cloud

Manifold cloud is an option for teams who want to run their policy on Manifold's GPUs. You wrap your policy and package it into a container. Manifold then runs that container on demand for each evaluation. If you would rather serve your policy from your own machine, see Self-hosted policy.

Start this process on the machine that stores your policy code and weights.

Requirements

• Policy code and weights
• The manifold CLI, authenticated with manifold auth login
• Docker
• Enough disk space to build your policy container locally
• A container registry like ghcr.io where you can push public images

Handing it off to your agent

Using agents

While this process can be done manually, we highly recommend letting your agent of choice handle it, with the help of our skills, which teach Codex, Claude Code, Cursor, and 70+ other AI agents to work with Manifold.

First, navigate to your project repository.

cd my-project

Then, install the Manifold agent skills.

npx skills add bifrostai/manifold-skills --all

From inside your policy repo, invoke the containerize-policy skill:

Claude

/containerize-policy

Codex

$containerize-policy

Your agent will ask a few quick questions about your policy, then walk you through wrapping, containerizing, and registering it on Manifold. Each step confirms with you before running the next.

The whole process takes a few minutes.

On completion

Once your image is public, your agent should register your policy with Manifold. If not, run:

manifold policy init POLICY \
  --image REGISTRY/NAMESPACE/IMAGE:VERSION \
  --minimum-gpu-memory-gb GPU_MEMORY_GB

The image tag becomes the version of your policy. Set --minimum-gpu-memory-gb to the GPU memory that your model needs, rounded up.

When your policy is successfully prepared, it will show up when you run:

manifold policy list