Porting an OpenClaw setup into Omniscio

What it is

Bring an existing OpenClaw gateway into Omniscio with as little typing as possible — first connect, then optionally have a Claude session recreate your setup (scheduled jobs and command/agent personas) inside Omniscio.

OpenClaw is a separate "gateway" server you run elsewhere (a home box, a cloud droplet). Omniscio talks to it over a WebSocket. To connect, Omniscio needs only two things: the gateway's address (ws://your-server:18789) and an auth token (a long hex string). The hard part has always been finding and re-typing those — this feature removes that friction and adds a one-click migration.

Where to find it

Where it lives

Settings → OpenClaw.

How it behaves

1. Connect with no typing

In the Bring in your setup box (shown when you're not connected), paste whatever you already have and Omniscio fills in the address + token for you:

• a ws:// or wss:// URL, or your Web UI address http://your-server:18789 (Omniscio converts http→ws and adds the default port :18789 automatically),
• a bare token (e.g. the contents of ~/.openclaw/gateway-token.txt),
• a chunk of your openclaw.json (Omniscio reads gateway.auth.token),
• your service env lines (CLAWDBOT_GATEWAY_TOKEN=…),
• or a mix of the above.

Prefer a file? Import from config file… opens a picker; point it at your openclaw.json / token file / .env and Omniscio extracts the same two values. Then press Connect.

(The parsing happens on your machine; the connection field only ever accepts ws:///wss://, so a stray http:// can't reach the network layer — Omniscio normalizes it for you first.)

2. Port your whole setup (optional, once connected)

Press Port my setup into Omniscio. Omniscio reads a snapshot of your gateway's setup and spins up a Claude session whose job is to recreate the equivalents in Omniscio:

In OpenClaw	Becomes in Omniscio
Scheduled jobs (cron)	Omniscio scheduled jobs — created switched off so you review them before they run
Commands & agent personas	Omniscio recipes (reusable prompts)
Custom model providers / API keys	Not copied — you get a checklist of keys to add yourself
Channels (Telegram, etc.)	No Omniscio equivalent — flagged as "not portable"

The session works through your setup and reports what it created, what it skipped (already existed), and what you need to set up by hand.

Why a Claude session does it

Translating OpenClaw concepts into Omniscio concepts takes judgment (what maps, what doesn't), so Omniscio hands the fuzzy part to a reasoning agent rather than a brittle rules table. The agent only ever proposes a structured plan — a dedicated Omniscio step is the one that actually creates the recipes and jobs, and it re-checks every item against Omniscio's own rules first (for example, a schedule that would run more often than every 5 minutes is rejected). So the agent can't make Omniscio do anything outside the allowed list, your jobs start switched off, and your secrets are stripped out before the session ever sees them.

Safety notes

• Nothing runs on its own. Ported scheduled jobs are created switched off — flip on the ones you want.
• Secrets stay put. API keys, the gateway token, and bot tokens are never included in what the session reads, never copied into a recipe, and never logged.
• Everything the porting creates is normal Omniscio data you can edit or delete.

For agents

For developers

The connection parser, the redacted snapshot builder, and the "applier" (the trust boundary that creates the artifacts) are covered by openclaw-port-contract.md. The spawned session hands its plan back through the control-server route POST /openclaw-port/apply.

Related

• openclaw-provider.md — the provider these ported settings run on.
• automation-credentials.md — where the keys a ported job needs are kept.
• cron-session-jobs.md — the scheduled jobs a port recreates.