Hermes - Ankra Platform

Overview

Hermes is Ankra’s autonomous infrastructure agent. It runs on the open-source OpenClaw runtime and connects to Ankra through the CLI, so it can manage clusters, build stacks, query Ankra AI, and provision the backing services your application needs — all from chat platforms like Slack, Discord, and Telegram. Hermes is built to be given real responsibility. You can hand it full control to design and configure the infrastructure your app depends on — a Postgres database, a Redis cache, an ingress, observability — and it will draft the architecture, produce a detailed bill of materials, and ask for your confirmation before it builds anything. This guide covers:

What Hermes can do — skills, the CLI, Ankra AI, stacks, profiles, and full architecture design
Deploying Hermes on any Kubernetes cluster as an Ankra stack
Connecting Hermes to Ankra by adding the Ankra CLI as a skill
The safeguards that keep autonomy safe — confirmations, drafted architecture, and a detailed bill of materials

Hermes runs on the OpenClaw runtime. If you are looking for the OpenClaw integration, this is now Hermes — the deployment mechanics are unchanged.

What Hermes can do

Skills

Extend Hermes with skills. The Ankra CLI skill turns natural language into cluster, stack, and credential operations.

Ankra CLI

Hermes drives the full ankra CLI — clusters, stacks, charts, credentials, and tokens — on your behalf.

Ankra AI

Hermes asks Ankra AI about live cluster state for health checks, root-cause analysis, and troubleshooting.

Build stacks

Hermes composes Helm addons and manifests into stacks and deploys them through Ankra’s GitOps flow.

Profiles

Reusable agent profiles bundle a model, skills, guardrails, and an autonomy level so Hermes behaves consistently per environment.

Design your app's tools

Hand Hermes a goal (“my app needs a database and a cache”) and it designs the architecture, lists the components, and builds them after you confirm.

Part 1: Deploy Hermes as an Ankra Stack

Deploy Hermes to any Kubernetes cluster managed by Ankra using the Stack Builder. This gives you a self-hosted, containerized agent running inside your own infrastructure.

Prerequisites

A cluster imported into Ankra with the agent connected
An API key from Anthropic or OpenAI
A Helm registry added for the OpenClaw chart repository (https://charts.openclaw.ai)

If you haven’t added the OpenClaw Helm registry yet, go to Settings → Registries → Add Registry and enter https://charts.openclaw.ai.

Step 1: Create the Stack

Open Stack Builder

Navigate to your cluster → Stacks → Create Stack.

Name Your Stack

Name it hermes or ai-agent.

Step 2: Add the Hermes Chart

Add the Chart

Click + Add → search for openclaw from the OpenClaw registry (this is the runtime Hermes is built on).

Configure Core Settings

Click the component and set these values:

replicaCount: 1

image:
  repository: ghcr.io/openclaw/openclaw
  tag: "latest"

config:
  model: "claude-sonnet-4-20250514"
  port: 8789
  logLevel: "info"

Hermes runs as a single instance — it does not support horizontal scaling. Keep replicaCount: 1.

Configure API Key

For quick setup, set the key directly:

config:
  anthropicApiKey: "sk-ant-..."

For production, use a Kubernetes Secret instead (see Production Setup below).

Configure Resources

resources:
  requests:
    memory: 512Mi
    cpu: 250m
  limits:
    memory: 1Gi
    cpu: 1000m

Enable Persistent Storage

Hermes stores workspace data and conversation memory. Enable persistence so this survives pod restarts:

persistence:
  enabled: true
  size: 5Gi

Encrypt sensitive values with SOPS: In the manifest edit view, click the SOPS button to encrypt your API key. This ensures the key is stored encrypted in your GitOps repository. See SOPS Encryption for setup instructions.

Step 3: Expose the Gateway (Optional)

If you want to access Hermes from outside the cluster, configure an ingress:

ingress:
  enabled: true
  className: "nginx"
  annotations:
    cert-manager.io/cluster-issuer: "letsencrypt-prod"
  hosts:
    - host: hermes.your-domain.com
      paths:
        - path: /
          pathType: Prefix
  tls:
    - secretName: hermes-tls
      hosts:
        - hermes.your-domain.com

Alternatively, use port-forwarding for local access:

kubectl port-forward svc/hermes 8789:8789 -n hermes

Step 4: Deploy

Review

Your stack should contain the chart with your configured values.

Save and Deploy

Click Save, then Deploy. Watch progress in Operations.

Verify

After 1-2 minutes, the Hermes pod should be running:

kubectl get pods -n hermes

NAME                      READY   STATUS    RESTARTS   AGE
hermes-6f8d9b7c4-x2k9p    1/1     Running   0          90s

Production API Key Management

For production deployments, store your API key in a Kubernetes Secret rather than in plain-text values:

Create the Secret

kubectl create secret generic hermes-api-key \
  --from-literal=anthropic-api-key="sk-ant-..." \
  -n hermes

Reference in Values

config:
  anthropicApiKey: ""

existingSecret:
  name: hermes-api-key
  anthropicApiKeyKey: anthropic-api-key

Security Hardening

Lock down the Hermes pod with security contexts:

podSecurityContext:
  runAsNonRoot: true
  runAsUser: 1000
  fsGroup: 1000

securityContext:
  allowPrivilegeEscalation: false
  readOnlyRootFilesystem: true
  capabilities:
    drop:
      - ALL

Hermes has shell access and can read files inside its container. Kubernetes provides meaningful isolation through container boundaries and network policies. For sensitive environments, apply a NetworkPolicy to restrict egress to only the AI provider API endpoints your model requires.

Network Policy Example

Restrict Hermes’ network access to only the AI provider API:

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: hermes-egress
  namespace: hermes
spec:
  podSelector:
    matchLabels:
      app.kubernetes.io/name: openclaw
  policyTypes:
    - Egress
  egress:
    - to:
        - ipBlock:
            cidr: 0.0.0.0/0
      ports:
        - protocol: TCP
          port: 443
    - to:
        - namespaceSelector: {}
      ports:
        - protocol: UDP
          port: 53

You can add this as a manifest in your stack alongside the chart.

Connect a Chat Platform

Once Hermes is running in your cluster, connect it to your team’s chat:

Platform	Configuration
Slack	Set `config.slackBotToken` and `config.slackAppToken` in your values, or use `existingSecret`
Discord	Set `config.discordToken` in your values
Telegram	Set `config.telegramToken` in your values

Store all chat platform tokens in a Kubernetes Secret and reference them via existingSecret for production use.

Part 2: Add Ankra CLI as a Hermes Skill

Once Hermes is running (either via the stack above or any other installation), you give it the ability to manage your Ankra infrastructure by adding the CLI as a skill.

Prerequisites

Hermes installed and running (or deployed as a stack above)
Ankra CLI installed and authenticated
An Ankra API token (for non-interactive auth)

Step 1: Install and Authenticate the Ankra CLI

If you haven’t already, install the Ankra CLI:

bash <(curl -sL https://github.com/ankraio/ankra-cli/releases/latest/download/install.sh)

Create an API token for Hermes to use (this avoids browser-based SSO, which doesn’t work in headless environments):

ankra login
ankra tokens create hermes-agent

Save the returned token — you’ll need it in the next step.

Step 2: Create the Ankra Skill

Create the skill directory and manifest:

mkdir -p ~/.openclaw/skills/ankra

Create ~/.openclaw/skills/ankra/SKILL.md with the following content:

---
name: ankra
version: 1.0.0
author: your-org
description: >
  Manage Kubernetes clusters and infrastructure on the Ankra platform.
  Use when the user wants to list clusters, deploy stacks, check cluster health,
  manage addons, search Helm charts, scale workers, design application
  infrastructure, or troubleshoot Kubernetes issues.
permissions:
  - shell
  - network
config:
  api_token:
    type: string
    required: true
    description: "Ankra API token for authentication"
    secret: true
---

# Ankra Platform Management

You have access to the `ankra` CLI to manage Kubernetes clusters on the Ankra platform.

## Available Commands

### Cluster Operations
- `ankra cluster list` -- List all clusters
- `ankra cluster get <name>` -- Get cluster details
- `ankra cluster select` -- Select a cluster (use `--name` for non-interactive)
- `ankra cluster reconcile [name]` -- Trigger reconciliation

### AI Chat
- `ankra chat "<question>"` -- Ask Ankra AI about your infrastructure
- `ankra chat health` -- Get cluster health summary
- `ankra chat health --ai` -- Get AI-analyzed cluster health

### Stack Management
- `ankra cluster stacks list` -- List stacks
- `ankra cluster stacks create <name>` -- Create a stack
- `ankra cluster stacks delete <name>` -- Delete a stack
- `ankra cluster stacks history <name>` -- View stack change history

### Helm Charts
- `ankra charts search <query>` -- Search for charts
- `ankra charts info <name>` -- Get chart details
- `ankra charts list` -- List available charts

### Hetzner Cluster Provisioning
- `ankra cluster hetzner create --name <n> --credential-id <id> --location <loc>` -- Create cluster
- `ankra cluster hetzner workers <id>` -- Get worker count
- `ankra cluster hetzner node-group list <id>` -- List node groups
- `ankra cluster hetzner node-group scale <id> <group> <count>` -- Scale a node group
- `ankra cluster hetzner deprovision <id>` -- Deprovision cluster

### OVH Cluster Provisioning
- `ankra cluster ovh create --name <n> --credential-id <id> --region <r>` -- Create cluster
- `ankra cluster ovh workers <id>` -- Get worker count
- `ankra cluster ovh scale <id> <count>` -- Scale workers

### UpCloud Cluster Provisioning
- `ankra cluster upcloud create --name <n> --credential-id <id> --zone <z>` -- Create cluster
- `ankra cluster upcloud workers <id>` -- Get worker count
- `ankra cluster upcloud scale <id> <count>` -- Scale workers

### Credentials
- `ankra credentials list` -- List all credentials
- `ankra credentials get <id>` -- Get credential details

### API Tokens
- `ankra tokens list` -- List API tokens
- `ankra tokens create <name>` -- Create a token
- `ankra tokens revoke <id>` -- Revoke a token

## Guardrails

- Always confirm with the user before running destructive operations like `deprovision`, `delete`, or `scale` (down).
- When scaling workers, show the current count first and ask for confirmation.
- Never expose API tokens or credentials in responses.
- For cluster creation, list the parameters back to the user and ask for confirmation before executing.
- Before building application infrastructure, present a drafted architecture and a bill of materials, then wait for explicit approval.
- Prefer `ankra chat` for troubleshooting questions -- it has access to live cluster data.

Step 3: Configure the Skill

Add the Ankra skill to your Hermes configuration at ~/.openclaw/openclaw.json:

{
  "skills": {
    "entries": {
      "ankra": {
        "enabled": true,
        "apiKey": "your-ankra-api-token",
        "env": {
          "ANKRA_API_TOKEN": "your-ankra-api-token"
        }
      }
    }
  }
}

Replace your-ankra-api-token with the token you created in Step 1.

If you already have other skills configured, just add the ankra entry to your existing entries object.

Step 4: Verify the Integration

Restart Hermes or wait for the skill watcher to pick up the changes, then test:

You: List my Ankra clusters
Hermes: Running `ankra cluster list`...

NAME              PROVIDER   STATE     WORKERS
staging-cluster   hetzner    running   3
production        ovh        running   5

Use cases

Use the Ankra CLI

With the Ankra CLI skill installed, Hermes turns plain language into CLI operations. You never have to remember command syntax — describe the outcome and Hermes runs the right commands.

You: Reconcile the production cluster
Hermes: Running `ankra cluster reconcile production`... reconciliation triggered.

Use Ankra AI

Hermes uses ankra chat to consult Ankra AI, which has live access to your cluster state. Use it for health checks, root-cause analysis, and “why is this broken” questions.

You: Why are pods crashing on production?
Hermes: Let me check with Ankra AI...
        Running `ankra chat "Why are pods in CrashLoopBackOff?"` on production...

        Ankra AI found 2 pods in CrashLoopBackOff in the api namespace:
        - api-server-7d8f9: OOMKilled (memory limit 256Mi, usage peaked at 312Mi)
        - api-worker-3k9d2: exit code 1, missing DATABASE_URL env var

Build stacks

Hermes composes Helm addons and manifests into stacks and deploys them through Ankra’s GitOps flow.

You: Create a monitoring stack on my staging cluster
Hermes: I'll create a new stack called "monitoring" on staging-cluster.
        Running `ankra cluster stacks create monitoring`... Done.
        I'll add kube-prometheus-stack and Loki. Deploy now? (y/n)

Profiles

A profile is a reusable agent configuration that bundles a model, a set of skills, guardrails, and an autonomy level. Profiles let Hermes behave consistently and appropriately per environment — a cautious, read-mostly profile for production and a freer profile for a sandbox. Define profiles in your Hermes configuration:

{
  "profiles": {
    "production": {
      "model": "claude-sonnet-4-20250514",
      "skills": ["ankra"],
      "autonomy": "confirm-all",
      "guardrails": {
        "require_confirmation": ["delete", "deprovision", "scale-down"],
        "require_architecture_draft": true,
        "require_bill_of_materials": true
      }
    },
    "sandbox": {
      "model": "claude-sonnet-4-20250514",
      "skills": ["ankra"],
      "autonomy": "auto",
      "guardrails": {
        "require_confirmation": ["deprovision"]
      }
    }
  }
}

You: Switch to the production profile
Hermes: Now using the "production" profile. I'll confirm every change and draft
        architecture with a bill of materials before building anything.

Give Hermes full control to design your app’s tools

This is what Hermes is built for. Hand it a goal and let it design and configure the infrastructure your application needs — a database, a cache, ingress, secrets, and observability — end to end. Hermes works in three phases: design, confirm, build. It never provisions anything before you approve the drafted architecture and its bill of materials.

You: My app needs a Postgres database and a Redis cache on staging. Set it up.

Hermes: Here's the architecture I propose for staging-cluster.

        Drafted architecture
        ─────────────────────
        app-data (stack)
          ├─ postgres        primary + 1 read replica, 20Gi PVC, daily backups
          ├─ redis           single master + 1 replica, 2Gi PVC, AOF persistence
          └─ secrets         SOPS-encrypted credentials in the GitOps repo

        Bill of materials
        ─────────────────────
        Component   Chart                       Version   Replicas   Storage   Est. resources
        Postgres    bitnami/postgresql-ha       14.x      2          20Gi      1 vCPU / 2Gi req
        Redis       bitnami/redis               19.x      2          2Gi       0.5 vCPU / 512Mi req
        Secrets     manifest (SOPS)             -         -          -         -

        Connection details will be written to a Kubernetes Secret named
        `app-data-credentials` and referenced by your app via existingSecret.

        Proceed to build this stack? (y/n)

You: y

Hermes: Building...
        Running `ankra cluster stacks create app-data`... Done.
        Adding postgresql-ha, redis, and the SOPS secret manifest...
        Saving and deploying. Watch progress in Operations.

Hermes can iterate on the design before building — ask it to add a read replica, change storage sizes, swap a chart, or pin versions, and it will re-draft the architecture and bill of materials before asking again.

Safeguards

Autonomy is only safe with rails. Hermes enforces these by default, and you can tighten them per profile.

Confirmations

Destructive and provisioning actions (delete, deprovision, scale-down, building infrastructure) require explicit approval before they run.

Drafted architecture

Before building anything, Hermes presents the architecture it intends to create so you can review and adjust it.

Detailed bill of materials

Every build is accompanied by a bill of materials — charts, versions, replicas, storage, and estimated resources — so there are no surprises.

Secrets stay secret

Credentials are written to Kubernetes Secrets and encrypted with SOPS in GitOps. Hermes never echoes tokens or credentials back in chat.

Match the autonomy level to the environment. Use a confirm-all profile with architecture drafts and a bill of materials required for production, and reserve freer profiles for sandbox clusters.

Sandboxed Environments

If you run Hermes in sandboxed mode (Docker), the ANKRA_API_TOKEN environment variable won’t be inherited automatically. Add it to your sandbox config:

{
  "agents": {
    "defaults": {
      "sandbox": {
        "docker": {
          "env": {
            "ANKRA_API_TOKEN": "your-ankra-api-token"
          }
        }
      }
    }
  }
}

You’ll also need to ensure the ankra binary is available inside the container. Either mount it or install it in a custom image:

FROM openclaw/sandbox:latest
RUN bash <(curl -sL https://github.com/ankraio/ankra-cli/releases/latest/download/install.sh)

Troubleshooting

Stack Deployment Issues

Issue	Solution
Pod stuck in `Pending`	Check for insufficient resources with `kubectl describe pod -n hermes`. Increase node capacity or reduce resource requests.
`CrashLoopBackOff`	Check logs with `kubectl logs -n hermes -l app.kubernetes.io/name=openclaw`. Usually a missing or invalid API key.
Ingress not working	Verify your ingress controller is installed and the `className` matches. Check cert-manager logs for TLS issues.
PVC not binding	Ensure a StorageClass exists. Run `kubectl get storageclass` to verify.
Helm chart not found	Confirm the OpenClaw registry (`https://charts.openclaw.ai`) is added under Settings → Registries.

Skill Integration Issues

Issue	Solution
”ankra: command not found”	Ensure the Ankra CLI is installed and in PATH. Run `which ankra` to verify.
”Unauthorized” errors	Check that `ANKRA_API_TOKEN` is set correctly. Re-create the token with `ankra tokens create`.
Skill not appearing	Verify the file is at `~/.openclaw/skills/ankra/SKILL.md` and the skill watcher is enabled.
Commands hang	The CLI may be waiting for interactive input. Ensure you’re using `--name` flags for non-interactive selection.
Sandbox can’t reach Ankra API	Add `network` to the sandbox’s allowed permissions and ensure DNS resolution works inside the container.

The Ankra CLI stores its config at ~/.ankra.yaml. When using API token auth via ANKRA_API_TOKEN, no config file is needed.

Next Steps

Stacks

Learn more about building and managing stacks in Ankra.

SOPS Encryption

Encrypt sensitive values in your stack configuration.

Ankra CLI

Full CLI reference for all Ankra commands.

Monitoring Stack

Add observability alongside your Hermes deployment.

​Overview

​What Hermes can do

Skills

Ankra CLI

Ankra AI

Build stacks

Profiles

Design your app's tools

​Part 1: Deploy Hermes as an Ankra Stack

​Prerequisites

​Step 1: Create the Stack

​Step 2: Add the Hermes Chart

​Step 3: Expose the Gateway (Optional)

​Step 4: Deploy

​Production API Key Management

​Security Hardening

​Network Policy Example

​Connect a Chat Platform

​Part 2: Add Ankra CLI as a Hermes Skill

​Prerequisites

​Step 1: Install and Authenticate the Ankra CLI

​Step 2: Create the Ankra Skill

​Step 3: Configure the Skill

​Step 4: Verify the Integration

​Use cases

​Use the Ankra CLI

​Use Ankra AI

​Build stacks

​Profiles

​Give Hermes full control to design your app’s tools

​Safeguards

Confirmations

Drafted architecture

Detailed bill of materials

Secrets stay secret

​Sandboxed Environments

​Troubleshooting

​Stack Deployment Issues

​Skill Integration Issues

​Next Steps

Stacks

SOPS Encryption

Ankra CLI

Monitoring Stack

Overview

What Hermes can do

Part 1: Deploy Hermes as an Ankra Stack

Prerequisites

Step 1: Create the Stack

Step 2: Add the Hermes Chart

Step 3: Expose the Gateway (Optional)

Step 4: Deploy

Production API Key Management

Security Hardening

Network Policy Example

Connect a Chat Platform

Part 2: Add Ankra CLI as a Hermes Skill

Prerequisites

Step 1: Install and Authenticate the Ankra CLI

Step 2: Create the Ankra Skill

Step 3: Configure the Skill

Step 4: Verify the Integration

Use cases

Use the Ankra CLI

Use Ankra AI

Build stacks

Profiles

Give Hermes full control to design your app’s tools

Safeguards

Sandboxed Environments

Troubleshooting

Stack Deployment Issues

Skill Integration Issues

Next Steps