AI assistant
The Job AI Assistant is a chat panel built into the Cronable console. Describe what you want in plain language — “create a job that posts our overnight numbers to Slack every morning at 9” — and it builds the job for you. It can also change and remove jobs, and answer questions about Cronable (“what’s a connector?”, “how do I get told when a job fails?”).
Open it with the round AI button — the gold spark in the bottom-right of the dashboard. It’s a Business-and-up feature; on a plan without it, the panel shows an upgrade card instead.
Chat to author. Canvas to operate. Talking to the assistant is a great way to create a job. The visual canvas with run history is where you live with it afterwards.
What it can and can’t do
The assistant is contained by design — it isn’t trusted to behave, it’s built so it can’t misbehave. The only actions it has are:
- Create, change and delete jobs — the same operations you’d do by hand in the Inspector.
- Answer questions from a bundled knowledge base (this documentation, in the product).
It has no ability to run a command, open a shell, browse the web, or read a single secret value. So the worst case — even if you paste something adversarial into the chat — is a job you can see on your canvas that you didn’t want, which you just delete. It can’t reach past your jobs.
Two more guarantees:
- It runs as you. Every change goes through the same role checks, the change-approval gate and the audit trail as an edit you made by hand — an assistant is never a way around them. If approval is on, its changes wait in the Approvals queue and it tells you so.
- It confirms destructive steps. Deleting a job asks you to confirm first.
When the assistant creates or edits a job, the node appears on your canvas live — no refresh — and opens in the Inspector so you can review or tweak it.
Set it up (once)
The first time anyone opens the assistant, a short “Set up the AI Assistant” step asks how it should connect to an AI model. You bring your own — nothing is routed through Cronable, and no AI call is billed by us:
- Claude (API key) or OpenAI (API key) — paste your key. It’s stored encrypted on your machine and never shown again. You can optionally pick a model (or type any model id your key can reach); leave it blank for the recommended default.
- Claude Code (installed) — no key needed. This reuses the
claudeCLI already signed in on the machine, run in a locked-down, jobs-only mode.
Press Test & save to confirm the connection. An admin only needs to do this once; after that everyone on the plan can use it. Change it later from the gear in the panel’s header.
Codex is intentionally not offered. The Codex CLI can’t be constrained to a jobs-only mode, so running it would break the “touches nothing but your jobs” guarantee — so Cronable won’t run it.
Drive it from claude.ai or ChatGPT instead
Prefer to work from an AI tool you already have open? Turn on the MCP server and an outside assistant — claude.ai or ChatGPT — can perform the same safe, jobs-only actions from its own chat. It’s off by default until you switch it on.
See also
- Simple & Advanced modes — the console the assistant builds jobs into
- MCP server — the same actions, driven from claude.ai or ChatGPT
- OpenClaw & Hermes — running and orchestrating your own agents
- Security — roles, approvals, secret handling and the on-prem trust model