Skip to content

✨ New Agent Assistant

The New Agent Assistant is the AIForged Assistant form focused on creating and planning a new Agent.

Instead of manually creating every service and connection from scratch, you can describe what you want in natural language and let the assistant build an initial plan for you.

What it helps with

The New Agent Assistant is designed to help you:

  • create a new Agent faster
  • choose a sensible starting structure
  • propose a service flow based on your description
  • suggest service settings and configuration direction
  • suggest document categories
  • suggest prompt and schema work for LLM-based services
  • review a generated blueprint before committing changes

Tip

The more specific your requirements are, the better the assistant can shape the initial Agent and flow.

Typical workflow

A typical New Agent Assistant flow looks like this:

  1. Open the new-agent experience.
  2. Describe the solution you want to build.
  3. Optionally start from one of the assistant’s quick-start ideas or templates.
  4. Let the assistant generate a proposed Agent blueprint.
  5. Review the blueprint preview.
  6. Adjust the proposal through follow-up prompts.
  7. Confirm and commit the blueprint when it looks correct.
  8. Validate the created Agent, Services, prompts, categories, and links.

What the blueprint usually covers

The assistant’s proposed blueprint can include:

  • Agent name and description
  • top-level services
  • pre-processing or post-processing utilities
  • child or dependency services where appropriate
  • service links and flow order
  • document categories
  • prompt and schema design work for LLM-based services
  • suggested service settings

This gives you a structured starting point before anything is finally applied.

What to tell the assistant

When asking the New Agent Assistant to build something, try to include:

  • the business goal
  • the document types involved
  • the input source
  • the target outputs
  • any required verification or HITL steps
  • any needed utilities
  • any external routing, exports, or webhooks

Example

Build an agent that reads invoices from email, classifies invoice vs statement, extracts invoice header fields and line items, validates supplier names against a dataset, and sends low-confidence items to verification.

Better example

Build an agent for AP invoice automation. Input comes from a shared mailbox. We need email scraping, attachment handling, invoice classification, invoice extraction, supplier lookup against a custom dataset, and a verification step for low-confidence totals or missing invoice numbers. The final result should be available in outbox and ready for export.

Quick-start templates and starter prompts

The New Agent Assistant can also help you get started with quick-start patterns instead of an empty prompt.

These are useful when:

  • you want a fast starting point
  • your use case matches a common intake/extraction pattern
  • you want to learn how AIForged composes a working flow

A quick-start path is usually best treated as a starting draft. Review the generated blueprint before committing it.

Prompt design during new-agent setup

When your new Agent includes LLM-based services, the New Agent Assistant can also coordinate early prompt work.

That means it can:

  • identify where prompt design is needed
  • queue or suggest prompt/schema design work
  • surface prompt design progress in the blueprint preview
  • hand prompt refinement off to the Prompt Designer Assistant workflow where needed

See: Prompt Designer Assistant

Commit and confirmation

A generated blueprint is not the same thing as a finished Agent.

Before changes are applied, review:

  • Agent metadata
  • service order
  • dependencies
  • categories
  • prompt design notes
  • settings that could materially affect behaviour

Warning

Treat the commit step as the moment where the proposed design becomes real configuration. Review carefully before committing.

After commit, validate the result in the normal Studio views.

After commit: what to check

Once the assistant has created or updated the Agent, check:

  1. the Agent card and Agent view
  2. the Service Flow Configurator
  3. service-specific settings
  4. parameter definitions and field structures
  5. prompt and schema content for LLM services
  6. document categories
  7. any required dataset links, utilities, or webhooks

See also: Service Flow Configurator

Permissions and context

The New Agent Assistant still operates within your normal AIForged access model.

That means:

  • your current Tenant matters
  • your current Group matters
  • your current roles still control what can actually be created or updated

If you can create a blueprint but cannot commit it, check:

  • whether you are in the correct Tenant
  • whether you have Owner, Administrator, or Developer access where required
  • whether related services or datasets are visible in the current context

Best practices

  • Start with a clear business outcome, not just a tool name.
  • Be specific about document types and desired outputs.
  • Use the blueprint as a collaboration aid, not as an unquestioned final design.
  • Validate prompt-heavy services before processing large volumes.
  • Process a small representative batch first.