Skip to main content
The Lab Agent is a conversational AI built into the Blueprint Editor. Instead of clicking through the canvas, you describe what you want in plain language and the agent builds or edits the logic for you — then runs your tests to check its own work. It is the fastest way to make a change when you can say what you want in a sentence. The agent works on the whole project: it can create, rename, and delete Blueprints, edit any section, build decision tables for rule-style logic, read and write your type definitions, and write and run test suites. It also builds new projects from a description — see Create a Project — and small apps for the Playground.

​
Opening the agent

Open the Agent tab in the right sidebar of the editor.
The Lab Agent panel: a request to raise a threshold, the agent's Read / Edit / Validate / Run steps and a reply referencing the Loan Application Scorecard Blueprint, with the composer and its Approval and Quality controls below

The Lab Agent works from a panel beside the canvas — here it takes a request, shows the steps it ran, and reports back referencing the Blueprint it changed.

Type your request in the composer at the bottom — “Add 15 points when the applicant has more than 5 years of employment” or “Explain what this Blueprint does” — and press Enter. The agent streams its response, showing its reasoning (“Thought for N seconds”) followed by the steps it takes. Click a step to see its details. Instead of typing, you can click the microphone (Start voice input) and speak your request.
The conversation is saved with the project — it reloads when you come back, and is scoped to the stream you are working on (main or a draft). Use Reset to start a fresh conversation. You can also revisit the conversation behind any past change from the Timeline with View conversation.

​
Point the agent at something

​
@-mentions

To point the agent at something specific, @-mention it. Type @ in the composer and pick a Blueprint, a node, an input or output, a type, or a comment — it is inserted as an inline pill. The agent also uses these pills in its replies, so there is no ambiguity about which element it means.

​
Point the agent at a node with Ctrl+I

Click a node, section, or element on the canvas and press Ctrl+I (Cmd+I on Mac). The node is dropped into the agent’s composer as a pill, so you can ask for a change to exactly that element — for example “round this to two decimals” or “add an else branch that declines” — without describing which one you mean.

​
Commands

Type / in the composer to see the commands: The command menu in the agent composer, listing /review — Review the project. Findings become comments — and /simplify — Simpler logic, same results. Can split blueprints. /review also checks that your descriptions still match the logic — use it before you share a project, so the specification your readers see matches what runs.

​
Choosing how it works

Two controls in the composer shape how the agent behaves: A context-usage indicator next to these controls shows how much of the conversation window is in use; long chats compact automatically so you can keep going.

​
Reviewing and applying changes

In Approval mode, the agent makes its edits in a draft and asks you to review them when it is done. The document shows what changed, the Diff Summary at the bottom lists every change from Main → Agent draft, and the agent’s message ends with Review agent changes: The editor after the agent raised the approval threshold — changed sections marked in the Contents sidebar and the document, the Diff Summary from Main to Agent draft with four changes, and the Review agent changes card with the Accept and Discard buttons
  • Click Accept to apply the changes.
  • Click Discard to drop them and leave your logic untouched.
Because the agent can also run your Blueprints and test suites, it usually verifies a change itself — for example “All 6 test cases in the test suite pass” — before handing it back to you.

​
Attachments

Click the paperclip (Attach file) to add a PDF or an image (PNG, JPEG, WebP) to a message — hand the agent a requirements document, a spec, or a screenshot of a rules table and ask it to turn it into logic.

​
Ask about Leapter itself

The agent can also answer questions about how Leapter works. It searches this documentation, reads the relevant pages, and links to them in its answer — for example “How do I make a Blueprint available through the REST API?”

​
Work with the project’s history

The agent can read the project’s Timeline: it lists changes and drafts, compares versions, and reports on reviews and releases. Ask it to “restore the version from before the threshold change”, and it restores that version — in Approval mode, you review the restore like any other change.

​
How the agent relates to other tools

​
Tips

  • Be specific. “Set the Approve threshold to 65 points” beats “make it stricter.”
  • Stay in Approval mode while you are learning what the agent does — you see every change before it lands.
  • Let it test. Ask the agent to run your test suite after a change so it catches its own mistakes.
  • @-mention to disambiguate when your project has several similarly named elements.
  • Run /review before you share a project, so descriptions and logic tell the same story.

​
What to do next

  • Test Suites — have the agent auto-generate test cases for your Blueprint
  • Playground — let the agent build an app to try your Blueprint
  • Test your Blueprint — run your logic and step through the execution