Agentic AI
An AI agent woven through the whole platform
Not a chatbot bolted on the side: an agent that plans multi-step work and carries it out — building forms, wiring logic, filing and finding submissions, navigating you to the result — inside the same product you use, with your permissions, showing every step it takes.
-
An agent in the app, not a chatbox
Ask for an outcome — “set up an intake form for equipment loans, with an approval step” — and the agent plans the steps, builds the draft, wires the logic and navigates you to the result. Every tool call it makes is visible in the conversation, so you watch the work happen rather than take its word for it. On a wide screen the panel docks beside the page, so what it is changing stays in view; it opens with three suggestions about the page you are on, and a file dropped anywhere on it is attached.
-
Import a PDF as a fillable form
Attach the paper form you use today; the agent reads it and produces a working, editable Faldaro draft — fields, types and structure — rather than a picture of one.
-
Prefill from a document
Upload a report, invoice or ID against a form and the matching fields fill themselves for your review. Only fields the form actually has; only values you approve.
-
Turn your paperwork into templates
Hand the document template editor the letter or certificate you send today and it comes back as a template — layout mirrored, logos extracted and placed, every fill-in area already a merge field for the form you picked. When it must match the original exactly, reproduce the PDF page for page instead and let one prompt bind your fields onto it. The notification editor does the same for email, subject line included. Or skip the upload and just describe the template you need.
-
Plain English to a filter
“Unpaid claims from July over $5,000” compiles into the exact filter — field keys, operators and workflow states all checked against the form before anything runs.
-
Translate a form for its submitters
Pick a language and the whole public form comes back translated — labels, help text, choices, the submit button — into the builder for you to read before you save. It fills gaps only: anything you worded yourself is never sent to the model and never overwritten, so running it again after you add a field translates the new strings and leaves your wording alone.
-
Style a field from a description
Tell the builder how a field should look and the model proposes the styling — theme colors, radius, scoped CSS — straight onto the canvas, validated by the same fence as hand-written styles, applied only when you save.
-
Reports and dashboard reads
Describe the report you need and refine the result in the report builder it seeds; ask the dashboard what changed this week and get an answer grounded in your own data.
-
Insight reports on any form
One click — or a weekly schedule — analyses a form’s submissions: trends, outliers, correlations, where work sits too long in the workflow. The statistics are computed by the server and the AI narrates them into ranked, actionable findings — and every count on a finding is the result of a real query, with “view affected submissions” opening exactly that list. Not a number the model made up: a number the database returned.
-
Predict outcomes from your own history
Train a model on past submissions — free-text answers included — to predict a field or the workflow outcome for the ones still open, with accuracy measured on held-out data and the driving fields named. Then it is held to account: as records close, live accuracy is measured against what actually happened, drift is said in words, and retraining is one click or a weekly schedule. Predictions appear as clearly labelled estimates beside a record with a what-if to test a change, never among its real values, and training costs no AI credits at all.
-
Flag the unusual ones
An anomaly model learns what a normal submission looks like and marks the ones that don’t fit — an “Unusual record” flag with the fields that made it odd — scored automatically as records arrive and change. Quiet by design: a flag on the record, not a siren in your inbox.
-
Standing agents that work a form’s submissions
Beyond the conversation: define an agent with a job in your own words — “triage new tickets and assign them” — against one form, with only the permissions you explicitly grant it (never more than your own). Run it on demand, on every new submission, or on a real schedule — every Monday at 9am in your timezone, not “roughly weekly, drifting later each run”. Every run is shown live, and a run that touches anything beyond its one form is refused.
-
Ask for a review
One click in the builder hands your draft to the agent, which walks the form, its logic and its workflow and reports what it finds — gaps, dead rules, unreachable states — as a visible investigation, not a verdict from a black box.
Annotation, not the record
AI output is checked before it touches anything
Every extraction, filter and generated definition is validated against your form's real fields before it is used — unknown fields are dropped, values are coerced to their declared types, and the server recomputes every calculation regardless of what was suggested.
Bounded by design
The agent is exactly as privileged as you
It signs its requests with your own session, so the platform's access control bounds everything it does. It has no delete tools, no admin tools and no credentials of its own — there is no 'AI user' with special powers to compromise.
Questions
AI, plainly
What makes the assistant agentic?
It runs a plan-and-act loop rather than generating text about your ask: it chooses from a set of tools — the same platform operations you use — executes them in sequence, checks the results and carries on until the job is done, reporting each step in the conversation as it goes.
What can the agent see and do?
Exactly what you can. It acts with your own permissions — every read and write it performs is bounded by the same access control as your own clicks, and it cannot delete or archive anything. Destructive actions stay in the UI, with you.
Does AI change my data without asking?
No. AI proposes and the human disposes: an extraction lands in the form for your review, a compiled filter is applied to your view, a generated report seeds the builder. Nothing persists until you take the normal save path.
What is an AI credit?
One unit of assistant or AI-builder work. Analysing a form’s submissions into an insight report costs about four; filling a form from a document costs about five, as does turning a typical page into a template — a dense multi-page reproduction spends more, at what it measurably cost; generating a whole form from a description costs about fifty. Training a predictive model costs no credits — credits meter model reasoning, not compute. Credits reset monthly, and stop at the allowance — a runaway loop cannot run up a bill. Paid plans can add prepaid credit packs that never expire; nothing is ever charged automatically.
Put the agent to work
Every plan includes monthly AI credits — the Free plan too. See pricing for allowances.
No card required.