Run Agents

Running an agent means deploying it to a scope, configuring what it should process, and executing the pipeline. Results appear in real time and can be saved automatically.

Deploy an agent

Before you can run an agent, deploy it to a scope:

  1. Go to the Tools page.
  2. Choose Global (available across all projects) or Current Project scope.
  3. Add an agent from your library or create a new one.

A deployment is a specific placement of an agent in a scope, with its own input configuration and settings.

Configure context

Each deployment has three context areas you can set up:

  • Input Sources — drag files, folders, or workspace documents from the sidebar. This is what the agent will process.
  • Reference (LLM Rules / Directives) — optional files that provide domain context, such as brand guidelines, templates, or style rules. These inform the AI without being the main content to process.
  • Output Folder — where saved output files will be written. Use the browse button or drag a folder from the navigator.

Run the agent

Click Run in the deployment inspector. You can also run agents from:

  • A pop-out Tool Runner window (Open in Window)
  • The ribbon (pin an agent for one-click access)
  • Mind map nodes
  • The automation canvas
  • The Chrome extension (with current page as context)
  • Triggers (scheduled or file-watch — see below)

What you see during execution

While an agent runs, you see:

  1. Step cards — one card per step with a status indicator (waiting, running, complete, or failed)
  2. Live streaming — AI steps display output token by token as it generates
  3. Progress — step count and elapsed time
  4. Final status — a completion or failure message when the pipeline finishes

Execution modes

Choose the execution mode in Agent Settings → Execution:

Autonomous mode (default)

The agent runs start to finish without pausing. AI steps always produce output without asking for more information. This is best for:

  • Automated triggers (scheduled runs, file watchers)
  • Running from the Chrome extension
  • Any workflow where human review is not needed mid-pipeline

Guided mode

After each AI step completes, a confirmation dialog appears showing the step output. You choose to:

  • Approve — continue to the next step
  • Reject — abort the remaining steps

Guided mode is useful when you want to review AI output before it gets saved or sent somewhere. It gives you a human checkpoint at each decision point in the pipeline.

Triggers

Instead of running agents manually, you can configure automatic triggers:

  • Schedule — run on a time-based schedule
  • File watch — run when a specific file or folder changes

Triggers use autonomous mode by default since there is no human present to approve guided steps.

Where results go

  • On screen — the last content-producing step is the primary visible output
  • Saved files — written to your output folder if saving is enabled (filename pattern: AgentName_timestamp.format)
  • External services — sent to connected APIs if the agent includes integration steps

You can control the save format (Markdown, text, JSON, CSV, HTML) and toggle file saving on or off in agent settings without rebuilding the agent.

First-run setup for API agents

If an agent uses external services (GitHub, Slack, ClickUp, etc.) and you have not configured credentials yet, the first run will prompt you to:

  1. Connect API credentials for the service
  2. Provide target parameters (such as a repository name or list ID)

These are saved to the agent settings for future runs.

Related