Magnets
Pin a model, instruction, Mind and automation to a single SmartPrompt — applied when it runs, handed back when it finishes.
What a Magnet is
A Magnet is a setting attached to a SmartPrompt that snaps into place when that prompt runs, then releases when it finishes. Four things can be pinned:
Without Magnets, a prompt that needs a big-context model and your legal Mind means three manual changes before you run it and three after. With Magnets, it means pressing one key.
Why they exist
Different jobs want different setups. Translating wants a fast cheap model and no Mind. Reviewing a contract wants a careful model, your legal Instruction and your clauses Mind. Logging a lead wants an automation.
Switching by hand between these is tedious enough that most people stop doing it and settle for one mediocre configuration. Magnets remove the switching, so each prompt runs in the setup that actually suits it.
Magnets pay off most in combination with a hotkey. One keystroke then selects the model, the instruction, the knowledge base, runs the prompt on your selected text, and pastes the result back.
Setting them up
How to reach it. Action grid → Prompts → edit a custom SmartPrompt → the Magnets section at the bottom.
Built-in prompts cannot carry Magnets. Clone one first if you want to pin settings to it.
Each lists what is available and active. Leave any at None to keep your current selection for that slot.
Magnets take effect immediately, including for a hotkey already assigned to that prompt.
The model dropdown lists every model from every connection, grouped by provider and sorted alphabetically, plus allowance models. Picking one records both the connection and the model, so the prompt always reaches the same endpoint.
The global Magnet switch
The magnet icon in the bottom bar — or Cmd+Shift+M — turns the whole mechanism on and off. On by default.
Switched off, prompts run with whatever you have selected in the top bar and their Magnets are ignored entirely. Useful for testing a prompt against a different model without editing it: turn Magnets off, choose the model manually, run.
Snap and release
The important behaviour: Magnets are temporary. Triplo records your current selections, applies the pinned ones, runs the prompt, and puts your selections back — whether the prompt succeeded or failed.
| Moment | Model | Instruction | Mind |
|---|---|---|---|
| Before | Allowance · Qwen3.7 Flash | Backend engineer | None |
| During the pinned prompt | OpenAI · gpt-5.6-sol | Contract review | Clause library |
| After | Allowance · Qwen3.7 Flash | Backend engineer | None |
You never have to remember to change anything back. That is what makes it safe to pin an expensive model to one prompt without it becoming your default by accident.
Partial pinning
Only the slots you set are affected. A prompt pinning just a Mind leaves your model and instruction exactly as they were. Pin the minimum the prompt actually needs.
Patterns worth copying
The cheap-and-fast prompt
Pin: a small fast model. Leave: everything else.
For mechanical work — fixing grammar, changing case, tidying a list — where a large model is a waste of money and latency. Give it a hotkey and it becomes a text tool.
The specialist prompt
Pin: a strong model, a matching Instruction, and the relevant Mind.
Contract review, code review against your architecture notes, medical summarisation. Everything the prompt needs to be good, travelling with the prompt.
The house-style prompt
Pin: an Instruction encoding your tone rules, and a Mind of approved examples.
Anything customer-facing where consistency matters more than flair. The Mind supplies previously approved wording; the Instruction enforces the rules.
The capture prompt
Pin: an Automation, plus a model that reliably emits JSON.
Turning selected text into a CRM record, a ticket, a calendar entry. Select the source material, press the hotkey, confirm, done.
The offline prompt
Pin: your local model.
For material that must not leave the machine. Pinning guarantees it, even if your default is a hosted model — no risk of forgetting to switch.
Limits & gotchas
- Custom prompts only. Built-in and shared prompts cannot carry Magnets — clone them first.
- The global switch wins. With the magnet icon off, every Magnet is ignored. If a pinned prompt seems to be using the wrong model, check that icon first.
- Deleted targets fail quietly. A Magnet pointing at a deleted connection, Instruction or Mind is skipped. If a pinned setup stops applying, re-open the prompt and check the dropdowns still show a value.
- Inactive targets are skipped. An Instruction, Mind or Automation toggled off will not apply even when pinned.
- Several prompts in one message: if you insert two chips that both pin a model, the last one wins. Prefer one pinned prompt per message.
- Magnets apply to background hotkey runs too, which is usually the point — but it means a prompt pinned to an expensive model will use it every time that key is pressed.
- Magnets are stored in the prompt's Markdown file under a
### Magnetsheading, referencing items by id. See File formats.