LoginDeck AI actions and using your own model
What spends an AI action, how many your plan includes, and how to point Flows at your own Anthropic, OpenAI-compatible or CLI model instead.
On this page
The AI parts of Flows — drafting one from a sentence, repairing a step that failed, writing a short comment — run on LoginDeck AI, which is included in your plan and needs no key. You can point them at your own model instead, and then nothing is metered by us.
Part of Pro, Team and Custom.
🤖 What spends an action#
One model call is one action.
| Feature | Actions it spends |
|---|---|
| ✨ Generate with AI — Don't look or Look at the start page only | 1, plus 1 if the draft needed an automatic correction round |
| ✨ Generate with AI — Walk the journey | 1 per page it reads, plus repairs and re-checks |
| ✨ Refine | 1 (plus 1 for a correction round) |
| ✨ Fix with AI / ✨ Fix in Run history | 1 (plus 1) |
| self-heal during a run | 1 per repair attempt |
| A stuck card | 1 for the card, and 1 more if your answer rewrites the Flow |
| The AI text step | 1 per call |
Templates never need AI. Running a Flow does not either, unless self-heal has to step in.
📊 What your plan includes#
| Plan | AI actions | Most in one day |
|---|---|---|
| Free | 20 in total, one time — they do not reset | – |
| Pro | 1,000 a month | 250 |
| Team | 2,500 a month | 625 |
| Custom | 5,000 a month | 1,250 |
The monthly allowance comes back on the 1st (UTC). The daily ceiling is a quarter of the month's allowance and resets at midnight UTC. Your Account page shows the meter — AI actions this month, or AI actions (free taste) on Free. Prices are on the pricing page.
In a team, members draw from the owner's allowance, and only if they have an edit or full grant across the whole workspace. See What members can see and do.
🔑 Use your own model instead#
Your own key or CLI is billed by that provider, not by us, and has no LoginDeck limits.
- Click ✨ Generate with AI on the Flows page, then the AI settings link at the bottom of the dialog. (It opens by itself if no model is set up.)
- Pick a Provider:
- LoginDeck AI — included in your plan (the default; nothing to set up)
- Anthropic (Claude) — your API key
- OpenAI-compatible — your API key
- Installed CLI — your subscription (Claude Code / Codex / Gemini)
- Fill in what that provider needs, then click Save.
| Field | When you need it |
|---|---|
| API key | Anthropic or OpenAI-compatible. Stored in this computer's keychain; it only ever leaves in the request to your provider. |
| Model | Optional. Defaults are claude-sonnet-5 for Anthropic and gpt-4o-mini for OpenAI-compatible. |
| Base URL | Only for a local model or a proxy — for example http://localhost:11434/v1 for Ollama. Pick OpenAI-compatible for those. |
| Which CLI + Command | For Installed CLI: Claude Code, OpenAI Codex, Gemini CLI or Custom command. Press Test to check it is installed and on this computer's PATH — you get ✓ Found — {version} or the error. |
Already paying for Claude Code, Codex or Gemini CLI?
Pick Installed CLI and LoginDeck runs the tool you have already signed into, so your existing subscription does the work. No key is stored. It is a few seconds slower than the API.
AI settings live on the computer you set them on — they are not synced. Set them on each machine you generate Flows from.
🚧 When you run out#
The messages say what to do:
- You have used the 20 AI actions that come with the free plan. Move to a paid plan for a monthly allowance, or add your own model key in AI settings.
- You have used all {n} AI actions in your {plan} plan this month. They come back on {date}. Move up a plan for more, or add your own model key in AI settings.
- That is {n} AI actions today, the daily ceiling on {plan} — it resets at midnight UTC. Add your own model key in AI settings to keep going now.
- too many AI actions at once — give it a minute — there is a limit of 40 a minute per account.
- AI actions in this workspace need an edit grant across the whole workspace — ask the owner — you are a team member without that grant.
- LoginDeck AI is not available right now — add your own model key in AI settings, or try again later
A call that fails upstream is not charged.
🔒 What the model sees#
- Pages the AI looks at are handled in memory on our server and never stored. We keep a count of actions, not your prompts or the replies.
- Values typed into a page are not part of what is sent — a password box's contents are never included.
- Saved logins and 2FA codes are stripped out of the variables handed to the model.
- With your own key or CLI, the request goes from your computer to your provider and we see none of it.
ℹ️ Good to know#
- Scheduled runs always self-heal, so they can spend actions unattended: Run Flows on a schedule.
- An AI text step can carry fallback lines, so a Flow still runs when no model is set up — it picks one of your lines instead. Leave the fallback box empty if you would rather the step failed.
- Tooltips in the Flow editor say "your AI model" even when LoginDeck's included model is doing the work.
- Turning self-heal off in the toolbar is the quickest way to stop a Flow spending actions while you are debugging it.