AI Builder
AI Builder turns a description of how your company works into a complete operational system: data model, screens, dashboard, business rules and roles. Your team owns it and keeps changing it by asking. After the first build, AI Builder drafts changes to schema, screens, rules and workflows against the current system as one connected plan, and the plan applies when an owner or admin approves it.
30-day trial with Pro capabilities · no card required
From a request to an applied change
An operations admin adds a client sign-off step for the delivery team. AI Builder reads the current system, proposes the connected changes across data, rules, screens and workflow, and waits for approval.
The request
“Deliverables need client approval before the team can mark them complete.”
Read-only investigation
Proposed plan 5 connected changes
Add client approval fields
Approved-by and approved-on fields on Deliverables
Block Completed without approval
A validation rule: Completed needs an approved-by value
Approval queue
Deliverables awaiting a client decision, grouped by project
Only account leads set the approval
A read-only rule on the approval field, with an exemption for the Account lead role
Hand off to the delivery lead
When the client approval is recorded, assign the next task
Applied in ordered stages, each verified after it lands. On Pro, build it in Development and publish it to Production as a release.
Diagram of the AI Builder approval path. The investigation, validation and fingerprint-bound approval follow the product's behavior; the table, screen and role names are sample data.
Want to see this on your own process? Our team will walk through it with you.
Talk to our teamWhat it does
Describe the operation in your own words. AI Builder asks clarifying questions, breaks the work into modules and builds the data model, screens, dashboard, business rules and roles as one connected system.
Ask for a new field, screen, rule, workflow step or dashboard as the operation changes. AI Builder works out what else the change touches and proposes it all together.
AI Builder reads the tables, screens, rules and workflows a change touches before it drafts the plan. When a change depends on existing records, it runs a read-only preparation step first.
A deterministic validator on the server checks every plan. Your approval is tied to that plan's fingerprint, so a changed proposal needs a new approval. Changes to schema, screens, rules and workflows need an owner or admin to request and approve them.
Changes to records arrive as a batch of up to 100 actions on one approval card. After writing, AI Builder reads each record back and compares it with what you approved.
Inside the apps it builds, people ask questions about their data and approve the record changes the AI drafts for them, within their own permissions.
The principle
After the first build, changes to schema, screens, rules and workflows arrive as one validated plan. The owner or admin who asked for it approves that exact plan, and a changed plan needs a new approval.
Under the hood
The mechanisms behind the page, for the architects and security reviewers evaluating it.
Read the technical documentation40 planning capabilities across 19 routes cover schema (tables, indexes, display, option lists), screens (create, edit, layout, navigation, actions), rule groups, automations, record changes, archived-value recovery, barcode templates and system-module installs into a Development environment (Pro).
Plans are validated by deterministic server code. Approval is bound to a fingerprint of the reviewed plan, so it cannot be reused for a different one, and only the person who started the chat answers its approval cards. Before a new task runs, an independent AI review checks the task definition.
Nine tools work outside the plan card, and this is the full list. On your request, set_table_parent nests a table under another, set_table_semantics edits table and column descriptions, patch_option_list edits a shared status or tag list, and ingest_file copies a file from a public link into your workspace storage. report_product_gap privately logs a request the product cannot meet, for Bondi's product team. cancel_workflow stops one of AI Builder's own running jobs and recover_archived_values fills empty cells from an archived column, each after you confirm; completed steps of a canceled job stay applied. install_system_module applies only the preview you confirmed, into a Development environment (Pro). ensure_barcode_template reuses an existing template or adds a new one to the approval of the screen that uses it.
AI Builder draws on the monthly AI credit included in every plan. The pricing page publishes the monthly AI credit for each plan and how Wallet balance covers additional usage.
Designed boundaries
The first build applies after its clarifying questions, without a plan card. Plan approval covers the changes that follow, so review the first version before your team relies on it.
AI Builder builds and changes the system. Automated AI work, such as document extraction and classification, runs as steps inside your workflows.
AI inside the workAI Builder does not build portals. Your team configures portals on top of the system it builds, on Pro.
Apps and portalsReview a plan the way you'd review a colleague's: with representative records and the affected roles in mind.
Availability
Questions buyers ask
It changes the environment you are working in. On Standard, yes: approved changes apply to the single Production environment. On Pro, build in Development and publish reviewed changes to Production; only owners and admins publish.
Nine named tools. On your request, AI Builder nests a table under another, edits table and column descriptions, edits a shared status or tag list, copies a file from a public link into your workspace, and privately reports a product gap to Bondi. It cancels one of its own running jobs or recovers archived values after you confirm, and installs a system module into Development (Pro) only from a preview you confirmed. A new barcode template joins the approval of the screen that uses it. Everything else AI Builder changes after the first build arrives as a validated plan or record batch for approval.
No. Your team configures portals on top of the system AI Builder builds. Portals are a Pro capability.
Owners and admins. Every AI Builder change to schema, screens, rules and workflows needs one of those roles, the person who asked approves the plan, and the approval is bound to that exact plan. Inside the apps, people approve their own record changes, within their own permissions.
Building with AI Builder uses your plan's monthly AI credit, the same credit your workflow AI steps draw on. Additional AI usage is paid from purchased Wallet balance, which does not expire.
Next step
Bring a process your company runs on. We'll walk through the system AI Builder builds for it and how your team keeps changing it. Or start a 30-day trial and describe it yourself.
