// help/ai-advisor/

The AI advisor.

Describe the job you want done in a sentence, and the advisor proposes a workflow for it. The model chooses and wires up the steps. ToolForte does the actual work, using the same capabilities you could have picked by hand.

What it is for

Knowing what you want is easy. Knowing which of our capabilities does it, and which field of one feeds which field of the next, is the fiddly part. That is the part the advisor takes.

You write what you want. It comes back with a proposed workflow: the steps in order, what each step is for in plain words, and the wiring between them already done. You read it, change anything you disagree with, test it, and save it. Nothing is saved until you say so.

You can start one from the advisor page, and take the proposal into the builder if you would rather finish it by hand.

How to describe a job well

The advisor is good at wiring and bad at guessing. Tell it the shape of the job and it will do the rest.

Name the input, the check and the result

A good description says what you will hand it, what should happen to that, and what you want back. It does not need to name our capabilities, and it should not name a technology: "a regular expression" is a solution, "every line that looks like an email address" is a job.

works well
  • Check whether an IBAN is valid and give me the bank country as a URL-safe slug.
  • Take a price without VAT and a rate, and give me the gross amount and the VAT amount separately.
  • Take a page title and give me a slug, a word count and a reading estimate.
too vague to build
  • Make my data better.
  • Automate my admin.
  • Do the thing I did last week but faster.

Keep it to a sentence or two; 1,000 characters is the ceiling. If the proposal is not what you meant, the fastest fix is usually to name the exact check or the exact output format you had in mind and ask again.

What it costs

One proposal costs 10 credits, the same as any other AI generation.

Per proposal10 credits, charged once, before the model is asked.
Internal retriesIf the first answer does not describe a valid workflow the advisor tries again, up to 2 attempts. That is still one charge, not two.
Running the resultSeparate, and priced per step like any workflow run. Proposing is not running.
If the model cannot be reached, the credits are still gone
They are taken before the request is sent, so nobody can spend our model budget for free. That means a failure on our side still costs you. The advisor says so on screen rather than hiding it, and support will put them back: get in touch if it keeps happening.

It only ever uses real ToolForte capabilities

The advisor picks from the 23 capabilities that actually exist. It cannot invent one, and it never computes an answer itself.

The division of labour is deliberate. The model understands what you asked for and chooses steps; the steps are then run by ToolForte, the same way a workflow you built by hand is run. An IBAN is checked by the same arithmetic either way, so the model has no opportunity to round the truth.

Every proposal is validated before you are shown it: the capabilities have to exist, the wiring has to point at steps that come earlier, and the whole thing has to fit inside the workflow limits, including the 12 step ceiling. A proposal that fails that check is not shown at all, not even partially, because a broken workflow on screen teaches you not to trust the feature.

It will tell you when your job cannot be done

An honest no is the point of the feature, not a failure of it.

If nothing in the catalogue comes close, the advisor says so, names what was missing, and builds nothing. If it can do part of the job it says that too: you get the workflow for the part that works, marked as partial, with the missing pieces listed by name so you know exactly what you still have to do elsewhere.

The gaps it names are useful to us as well. If you tell us what you needed, it becomes a candidate for the next capability we add.

Code it suggests is for your machine, never ours

Sometimes the honest answer to a gap is a few lines of code. The advisor may suggest some, clearly labelled.

ToolForte does not run suggested code
ToolForte does not run this code. It is a suggestion for your own environment, written by an AI model and not reviewed by us. Read it before you run it.

A snippet is a suggestion you can take to your own environment, and it comes with that same warning attached to it wherever it appears. Running model-written code on our servers would be exactly as dangerous as it sounds, so we do not, and no wording in a prompt changes that.

After the proposal

Read it before you keep it.

Each step comes with a line saying why it is there, so you can judge the workflow instead of trusting it. Change a step, add one, or rewire a field, then test it with a real input. The builder tells you what that test run costs before you press it. Save it when it does what you meant, and it joins your workflows like any other. How workflows work covers the rest.

Keep reading
Start here
What ToolForte is, and the three ways to use it: in the browser, over the API, and from an AI assistant.
Credits and plans
What a credit is, what each action costs, what resets monthly, and what happens when you run out.
Workflows
Chain capabilities into one job. Steps, inputs, test runs, run history, and how a run behaves when a check fails.
API and MCP
For developers: get a key, send the header, call a capability, run a workflow.
Your account
Signing up, confirming your email, resetting your password, usage history, billing and API keys.