SYS.DOCS // DOCS

OpenClaw

OpenClaw is an open-source AI agent gateway with a browser Control UI, chat channel integrations, skills, and persistent memory. Edka installs it from the agent catalog and runs it in your own Kubernetes cluster, in your own cloud account, backed by a persistent state volume.

It is one of three managed agent runtimes in Edka, alongside the Hermes Agent and the Codex Agent. OpenClaw is the channel-focused one: a single always-on assistant you reach from the Control UI and from chat platforms such as Telegram, Discord, Slack, or WhatsApp.

Edka ships its own OpenClaw image with the catalog plugin set compiled in, including the Prometheus exporter and web search, so the pod starts without downloading anything.

OpenClaw is available on every plan, including Free.

Requirements

A cluster you have write access to.
An API key for a model provider. OpenRouter, Anthropic, OpenAI, and Google Gemini are selectable at install; the key is stored as a Kubernetes Secret in your cluster.
For Control UI access from outside the cluster, a hostname configured under the cluster’s Domains. The Control UI can run sessions and manage credentials, so expose it behind a private boundary such as Tailscale or Cloudflare Zero Trust rather than on the public internet.
For state backups, an S3 compatible object storage integration. See Object Storage.

Deploy OpenClaw

Open a cluster, then go to Workloads → Agents.
Select Deploy Agent and choose OpenClaw.
Under Configuration, pick the model provider, paste its API key, and set the agent instructions.
Decide the optional features. All of them default to off:
- Enable Cron turns on OpenClaw scheduled tasks.
- Enable Dreaming runs nightly memory consolidation with scheduled background model calls billed to your API key.
- Enable Web Search gives the agent a web search tool through the Perplexity plugin, used with your API key.
- Enable Browser switches to the image variant with bundled Chromium and gives the agent a browser tool.
Complete the Storage, Resources, and Access tabs as needed.
Submit and watch the agent card for progress.

The agent card then shows runtime health, version, namespace, CPU and memory usage, and the latest state backup. Chart and image updates are tracked by the catalog, so the agent can be updated in place from its card.

Where settings live

OpenClaw manages its own configuration. Settings marked applied when the app is installed, such as the model, instructions, and the feature toggles, seed the initial configuration; after that, manage models, channels, skills, and schedules directly in the OpenClaw Control UI or CLI, and they persist across restarts and updates.

Edka keeps owning the infrastructure: resources, storage, networking and exposure, metrics, update policy, and session disk limits. Those follow the app settings on every deploy.

Updating or redeploying the agent keeps its state volume, including sessions, memory, channel logins, and pairing approvals. Reinstalling the app deletes the state volume. Take a state backup before removing an agent you may want back.

Control UI and browser pairing

Open the Control UI with Dashboard on the agent detail page. OpenClaw asks each new browser for a one-time device approval:

Open the dashboard and wait for the Control UI to show that pairing is required.
On the agent’s Devices tab, select Approve Latest. Edka approves the newest pending request against the running gateway; paste a request ID first to approve a specific device instead.
Retry the Control UI connection.

The Skip Browser Approval setting disables this check. Use it only when the Control UI is reachable exclusively through a trusted private boundary.

Channels and DM pairing

Connect chat channels in the OpenClaw Control UI under Settings → Channels. Token-based channels such as Telegram, Discord, Slack, Matrix, and Mattermost need only their credentials; WhatsApp is linked by scanning a QR code from the Control UI.

By default, unknown senders who message the agent receive a pairing code instead of a reply. Approve them from the agent’s Devices tab: enter the channel, select List Pending to see waiting requests, then enter the sender’s code and select Approve Code.

Usage and metrics

The Usage card on the agent overview shows model spend and token totals for the last 30 days, read from OpenClaw’s own accounting.

Enable Metrics on the app to chart model calls, tokens, cost, tool executions, runs, queue depth, and memory over time from the Metrics tab. A fresh agent shows zeros until it handles its first messages.

State backups

State backups export the OpenClaw state archive: sessions, memory, workspace files, skills, and channel logins. Enable them in the agent’s Backup configuration tab by selecting an S3 compatible object storage integration, a schedule, and a retention period. The archive includes channel credentials, so it always lands on your own storage.

The Backups tab runs scheduled backups from an in-cluster CronJob, takes manual Back Up Now runs, and lists stored backups. Backups run while the agent stays up and survive uninstall for their retention window.

Restoring is a manual operation: download the archive from your object storage and follow the OpenClaw backup restore guide to stage it into a fresh state directory.

Diagnostics

The Diagnostics card on the agent overview downloads a sanitized, payload-free support bundle with config, status, health, and redacted logs. Attach it when reporting OpenClaw issues; it contains no conversation content or credentials.