I am working on an open source project called Ridge that gives coding agents a common interface to resources such as local projects, Docker containers, SSH machines, and S3 buckets. Agents can discover available resources, access data, copy between them, and execute commands where supported. It works through MCP, a CLI, or Python.

I started building it because I wanted to use Codex on my laptop for kernel optimization experiments, with execution on my GPU box and models stored in S3. Before getting to the actual experiment, Codex needed to put together the file transfer and remote execution logic. You can save those scripts and leave instructions for the next session, but then you have a small infrastructure project to maintain alongside the experiment.

The idea is to make these operations reusable across projects and environments. For example, an agent can copy a model from S3 to a GPU machine in one operation, without passing the file contents through its context. The initial providers implement common capability contracts that can be extended to other resources. Ridge also supports giving subagents access to selected resources and operations, tracking background jobs, and coordinating shared file updates. The agent harness still handles planning and launching workers.

What do you think? Have you tried similar workflows with coding agents? I would be interested in hearing what you were trying to do, what got complicated, and what you ended up building or using to make it work.