jyje/hermes-agent-helm

👩🏻‍💻 A Helm chart to run Hermes Agent on Kubernetes, community-powered, lightweight

English · 한국어 · Chart docs · CONTRIBUTING · SECURITY · AGENTS

---

Found this useful? Please give it a ⭐ — it helps others find it.

Summary

Run Hermes Agent on Kubernetes with one helm install — works with any LLM provider Hermes supports, scales down to a single small node, and is verified to actually run, not just render. A community-powered chart, not an official Nous Research release.

Getting started

Add the Helm repository (published to GitHub Pages) and install:

helm repo add hermes-agent https://jyje.github.io/hermes-agent-helm
helm repo update
helm install hermes-agent hermes-agent/hermes-agent \
  --namespace hermes-agent --create-namespace \
  --set-string env.OPENAI_API_KEY='sk-...' \
  --wait

Each release is also published as an OCI artifact, so you can install directly from the registry without helm repo add:

helm install hermes-agent oci://ghcr.io/jyje/hermes-agent-helm/hermes-agent \
  --namespace hermes-agent --create-namespace \
  --set-string env.OPENAI_API_KEY='sk-...' \
  --wait

Optionally pin --version to a specific released chart version instead of latest.

To install from this repo's source instead (e.g. to try an unreleased change), see Quick start below.

Why this chart

• All of Hermes's provider support, via values.yaml. Hermes itself already supports openai-api, anthropic, gemini, openrouter, nvidia, deepseek, and any OpenAI-compatible endpoint (e.g. LiteLLM) through environment variables — this chart just exposes that config through values.yaml and ships ready-to-adapt examples per provider, with no provider baked into the templates.
• Lightweight → production. Sized for homelab / single-node / edge clusters out of the box (one replica, modest requests, a small PVC), and ready for production by scaling up — not out. Hermes is a single-instance personal agent (one HERMES_HOME, one gateway, one memory), so you don't replicate a pod; you run several well-managed instances and group them into a team that shares context over a common gateway channel. See Hermes teams.
• Verified end-to-end. CI installs the chart on an ephemeral kind cluster and runs the bundled test Job (hermes doctor). When the NVIDIA_API_KEY repository secret is available, it also runs a live hermes chat round-trip against NVIDIA NIM — not a mock. The Discord-thread leader team has also completed a live human → leader → two members → leader round-trip; Telegram team orchestration remains separate.

Deployment evidence: leader august and members may/march running as independent releases on kind. This screenshot does not by itself prove the multi-turn mention loop; see Hermes teams for the current status.

For the full resource breakdown, configuration model, and provider-by-provider install examples (including messenger integrations), see charts/hermes-agent/README.md.

Repository layout

charts/hermes-agent/                     # the Helm chart (see its README for the full values table)
charts/hermes-agent/values-*.yaml        # ready-to-adapt examples: providers, Discord/Telegram, LiteLLM (see chart README "More examples")
examples/helm/                           # install from Git and from OCI (ghcr.io) + publish guide
examples/argocd/                         # ArgoCD Application examples (one per values-*.yaml) + GitOps/SealedSecret pattern
charts/hermes-operator/                  # ⏸️ long-term, not started: possible install chart for an Agent/AgentTeam-CRD operator — placeholder, TBA (see Roadmap)
docs/                                    # deeper guides: teams.md (group instances into a team), collaboration.md (make them hand off by @mention), roadmap.md
.github/workflows/                       # ci (lint + docs-drift + real round-trip on kind) and release (version bump -> tag -> ghcr OCI)
CONTRIBUTING.md                          # branch model (dev/main + tags) + release-on-version-bump
AGENTS.md                                # design principles & workflow for contributors
Makefile                                 # docs / lint / template / install / test / package / push

Quick start

# render & lint
make template
make lint

# install with the generic defaults (set your provider key)
# release name == chart name keeps resources clean (hermes-agent-0, not hermes-agent-hermes-agent-0)
helm upgrade --install hermes-agent ./charts/hermes-agent \
  --namespace hermes-agent --create-namespace \
  --set-string env.OPENAI_API_KEY='sk-...' --wait

# run the install test (doctor-style Job)
helm test hermes-agent -n hermes-agent
kubectl logs -n hermes-agent -l app.kubernetes.io/component=test --tail=-1

# or start from a ready-made example (provider, Discord/Telegram, LiteLLM, ...)
helm upgrade --install hermes-agent ./charts/hermes-agent \
  --namespace hermes-agent --create-namespace \
  -f charts/hermes-agent/values-anthropic-and-discord.yaml \
  --set-string env.ANTHROPIC_API_KEY='sk-ant-...' \
  --set-string env.DISCORD_BOT_TOKEN='...' --wait

See charts/hermes-agent/README.md for the full values table, the "More examples" table (values-*.yaml for every supported provider plus Discord/Telegram and LiteLLM), and an ArgoCD example.

Development

Branch model, release process, and local checks (make lint / make docs / make test) are covered in CONTRIBUTING.md; chart design principles are in AGENTS.md.

CI/CD

• Every PR and every push to dev/main runs validate-chart.yaml: helm lint, helm template, a chart-docs drift check, and a full install + test on an ephemeral kind cluster (real hermes chat round-trip when an NVIDIA_API_KEY secret is available).
• Releases are Changesets-driven, not tag-push-driven. A user-visible chart change adds a patch, minor, or major entry under .changeset/. When you are ready to release, manually run propose-release.yaml to combine pending entries into one reviewable release PR, write its CHANGELOG.md notes, and synchronizes the private release manifest with Chart.yaml, Artifact Hub metadata, chart docs, and versioned examples. Review and merge that PR; then release-chart.yaml tags vX.Y.Z, writes the GitHub Release, and publishes the chart to oci://ghcr.io/<owner>/hermes-agent-helm/hermes-agent.

So: lint + test gate every change; the release itself is just a normal reviewed PR (the version bump) — the pending Changesets decide its SemVer, merging is what ships. See CONTRIBUTING.md for the full release playbook.

Roadmap

This chart deploys and manages one agent well; teams via an ArgoCD ApplicationSet are how you scale today, and a CRD-based operator is a long-term, not-started candidate. See docs/reference/roadmap.md.

Contributing

Issues, PRs, and ideas are all welcome — start with CONTRIBUTING.md (branch model, local checks, release flow). Every merged contribution is credited in the changelog and release notes.

Thanks to everyone who has contributed to and starred this project ⭐

---

Banner © Nous Research (MIT).