Hermes Atlas
Multi-agent & orchestration

Hermes Remote Pair

IShenH/hermes-remote-pair

Pair of skills that let any HTTP-capable agent delegate tasks to a remote Hermes Agent over its API

In short

Hermes Remote Pair is a pair of skill files that lets any AI agent able to send HTTP requests call a Hermes Agent running on another machine through its OpenAI-compatible API.

What Hermes Remote Pair does

The repository has two matching skills. hermes-remote-client/SKILL.md goes to the calling agent and covers when to delegate, how to assemble a task, synchronous versus asynchronous strategy, SSE progress listening, and error retry and fallback. hermes-remote-server/SKILL.md is for the administrator of the Hermes machine and covers enabling api_server, authentication keys, an nginx reverse proxy with CDN or WAF protection, the contract implementation and monitoring. Both share one caller contract with a task envelope, a response envelope, a state machine and SSE event names.

The envelope carries a task_id for idempotent retries, an untrusted_data field that keeps content from web pages, comments or logs marked as data rather than instructions, and an expected_output format. Short tasks use chat/completions synchronously, and because the README says free-tier CDN connections are cut at about 100 seconds, tasks over 90 seconds should use the asynchronous /runs endpoint with SSE progress. It presents itself as a lightweight point-to-point alternative to A2A with no SDK. The README is in Chinese and the license is MIT.

Key features

  • Client skill for the calling agent and server skill for the Hermes administrator
  • Task envelope with task_id, untrusted_data and expected_output fields
  • Synchronous calls for short tasks and asynchronous /runs with SSE for long ones
  • nginx reverse proxy guidance, including SSE buffering and CDN timeouts
  • A curl-based client that needs no SDK

When to use it

  • Letting a daily assistant agent delegate server work to a Hermes Agent that has terminal and database access
  • Running long Hermes tasks remotely with streamed progress events
  • Keeping content from web pages or logs marked as untrusted data when passing tasks between agents

Who it is for: People who run Hermes Agent on a server and want another agent to call it over HTTP in a controlled way.

How it fits with Hermes Agent

Built for Hermes Agent: the server skill configures the Hermes api_server platform, and the client skill lets other agents call it.

Requirements: Hermes Agent with platforms.api_server enabled and an API key; for public exposure an nginx reverse proxy with CDN or WAF protection; the caller needs only curl or another HTTP client

FAQ

What is Hermes Remote Pair?

Hermes Remote Pair is a matched pair of skills that let one agent delegate tasks to a Hermes Agent on another machine over an OpenAI-compatible HTTP API, with a shared task and response contract.

Does Hermes Remote Pair need an SDK?

No. The calling agent only needs to send HTTP requests, for example with curl, to the Hermes OpenAI-compatible endpoint with a bearer API key.

Is it safe to expose the Hermes API publicly?

The README warns that a leaked API key means a compromised server, because Hermes has terminal access. It advises strong random keys of at least 32 characters, environment variable storage, and a reverse proxy with CDN or WAF protection and a hard-to-guess path prefix.

Similar multi-agent for Hermes Agent

All multi-agent

Related guides: Build agent teams with the Hermes kanban board · Run multiple Hermes agents with profiles · Connect Hermes agents on several machines with Hermes Desktop