Hermes Atlas
Developer tools & SDKs · works with Hermes Agent

Remote Agent Server

ma-pony/remote-agent-server

Self-hosted task API and web console for running Claude Code, Codex and Hermes agents

In short

Remote Agent Server is a self-hosted service that connects command-line agents, including Claude Code, Codex and Hermes, to ticketing, CI and internal business systems. It provides an asynchronous task API, a web console, isolated per-session workspaces and signed webhook results.

What Remote Agent Server does

Deployed on your own machine, Remote Agent Server lets other systems submit work over an HTTP API, or lets people run the same agents from a web console. It handles queuing, workspaces, sessions and execution records, and returns progress through status queries, resumable SSE streams or signed webhooks. Submitted tasks keep running if the caller disconnects, and idempotency keys prevent duplicate runs. It talks to agents through acpx and the Agent Client Protocol and reaches external tools through MCP.

Prepared project environments are copied into a separate copy-on-write workspace for each session, using APFS clones on macOS and Btrfs snapshots on Linux. Administrators can choose which MCP tools each agent exposes, install and roll back Skills from Git repositories, set model policies and receive GitHub or GitLab webhook events. A single Fastify process with SQLite in WAL mode means no separate database or message queue. The README states that workspaces are not a container or a security sandbox.

Key features

  • Asynchronous Task API with idempotency keys and multi-turn conversations
  • Event queries, resumable SSE and signed webhooks for results
  • Per-agent selection of exposed MCP tools
  • Skills installed from Git with file previews and rollback
  • Copy-on-write workspace per session on APFS or Btrfs
  • GitHub and GitLab webhook event intake with filters

When to use it

  • Letting a ticketing or operations platform dispatch code investigation tasks
  • Calling an agent from a shell script or CI pipeline and collecting the result by callback
  • Running and continuing agent tasks from a browser console
  • Sending pull request or merge request events to a review agent

Who it is for: Developers and teams who already use agent command-line tools and want to wire them into long-running business workflows.

How it fits with Hermes Agent

Hermes is named alongside Claude Code and Codex as one of the agents it can run, so it is supported rather than the sole target.

Requirements: Node.js 22, pnpm 10 via Corepack, Git, at least one logged-in Provider CLI, and macOS with APFS or Linux with Btrfs (btrfs-progs)

Note: Workspaces are copy-on-write file copies, not containers or a security sandbox, and the service is aimed at single-host deployment for trusted users.

FAQ

What is Remote Agent Server?

Remote Agent Server is a self-hosted service that runs command-line coding agents on behalf of other systems. It adds a task API, a web console, isolated workspaces and signed webhooks.

Does Remote Agent Server work with Hermes Agent?

Yes. Hermes is named with Claude Code and Codex as an agent that can be connected, provided at least one Provider CLI is installed and logged in.

What do I need to run Remote Agent Server?

You need Node.js 22, pnpm 10 through Corepack, Git and a logged-in Provider CLI. Workspaces use APFS on macOS or Btrfs on Linux, which also needs btrfs-progs.

Similar dev tools for Hermes Agent

All dev tools

Related guides: How to install Hermes Agent · Run multiple Hermes agents with profiles