Hermes Concurrent Agents: multi-agent tool for Hermes Agent

Run a bounded, supervised team of Hermes Agent workers on one host, coordinated through Kanban

What Hermes Concurrent Agents does

HCA builds on Hermes Agent's Kanban board, profiles, sessions, workspaces and plugins. You initialise bounded worker profiles from an existing Hermes profile with hca init, then start a mission with hca run. One-step work uses a single worker. Fan-out happens only when a run has several acceptance criteria plus the --independent-criteria flag, and concurrency stays limited by concrete worker slots and admission capacity.

The tool adds pre-claim admission, exact process ownership, restart reconciliation and bounded review and rework, then produces one evidence-backed result with hca collect. The same versioned FleetService backs the human CLI and a set of Hermes plugin tools such as hca_team_run and hca_team_status. Optional presets read GB10 memory pressure and vLLM or SGLang metrics to make conservative admission decisions, while generic Linux operation works without CUDA or telemetry. HCA does not provision models, copy provider credentials or replace Hermes tools.

Key features

Pre-claim admission and concrete worker slots that cap concurrency
Exact process ownership with restart reconciliation
Bounded review and rework before one evidence-backed result
Hermes plugin tools including hca_team_run, hca_team_respond and hca_team_collect
Optional GB10 admission presets with vLLM and SGLang telemetry adapters
Approval-gated cancellation through hca stop

When to use it

Splitting a release-readiness report into independent research and audit tasks handled by parallel workers
Running a bounded implementation task and having a reviewer check it before the result is collected
Using a GB10 or DGX Spark machine to run several Hermes workers against one model endpoint

Who it is for: Developers who already run Hermes Agent on a Linux host and want several supervised workers pursuing one goal.

How it fits with Hermes Agent

Built around Hermes Agent: it relies on Hermes Kanban, profiles, sessions and plugins, and needs a configured Hermes install (0.18.2 / 2026.7.7.2 for the required stable contract lane).

How to install Hermes Concurrent Agents

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

git clone https://github.com/r0b0tlab/hermes-concurrent-agents.git
cd hermes-concurrent-agents
python3 -m venv .venv && . .venv/bin/activate
python -m pip install -e .

Requirements: Linux, Python 3.11 or 3.12, tmux, and Hermes Agent 0.18.2 / 2026.7.7.2 already configured and authenticated

Note: The README labels the project alpha with a single-host control plane, and remote agent placement is unsupported.

FAQ

What is Hermes Concurrent Agents?

Hermes Concurrent Agents (HCA) is a Python CLI and Hermes plugin that runs a small, bounded team of Hermes Agent workers on one goal. It coordinates them through Hermes Kanban and returns one evidence-backed result.

Does Hermes Concurrent Agents work with Hermes Agent?

Yes, it is built on Hermes Agent Kanban, profiles, sessions and plugins. The Kanban board and the workers must stay on one host, although a model endpoint can be remote through an ordinary Hermes profile.

What do I need to run Hermes Concurrent Agents?

You need Linux with Python 3.11 or 3.12, tmux, and a configured and authenticated Hermes Agent on PATH. CUDA, NVML and endpoint telemetry are optional.