Agent Mode

Agent Mode connects your terminal session to an AI model (Claude, GPT-4, Gemini, or a local Ollama instance). You give the agent a plain-language instruction and it reads the current screen, decides which keys to press, and executes them automatically, one step at a time, with full visibility.

Prerequisite: You need an API key from an LLM provider (Anthropic, OpenAI, Google) or a locally running Ollama instance. Ask your administrator whether a server-side key is already configured.

Connect to your session

Agent Mode works on any connected session. Connect to your AS/400, z/OS, or Linux system as described in the Getting Started tutorial.

What you should see: The terminal shows your host's initial screen (Sign On, ISPF, or shell prompt).

Open Agent Mode

Click Tools > Agent Mode or press Ctrl+Shift+A. The Agent panel slides in from the right.

What you should see: The terminal shrinks to the left and an Agent panel appears on the right with a text input field, a provider selector, and a run button. The terminal screen is still live.

Choose a provider and enter your API key

Select your LLM provider from the dropdown (Claude, GPT-4, Gemini, Ollama). If the key is not pre-configured by your admin, enter it in the API Key field. Keys are stored in your browser's local storage and never sent to Infomanta's servers.

What you should see: The provider dropdown shows the selected provider name. If you typed an API key, the field shows dots (masked). The Run button becomes active.

Type a natural-language instruction

Click the text area and describe what you want the agent to do. Be specific, include the screen name, field names, or values where you know them.

Sign on as DEMO with password DEMO and navigate to Customer Maintenance

Find the order with order number 12345 and change the status to Shipped

Run WRKACTJOB and filter by subsystem QINTER

What you should see: Your instruction appears in the text area. The character count updates. Press Ctrl+Enter or click Run to start the agent.

Watch the agent execute

The agent reads the current screen, sends it to the LLM, receives the next action (e.g. "type DEMO in field at row 7 col 23"), executes it, reads the updated screen, and repeats. Each step is logged in the Agent Log panel.

What you should see: The terminal updates automatically as the agent types and presses keys. The Agent Log shows each step: "Reading screen", "Typing username", "Pressing Enter", etc. A spinner indicates the agent is waiting for the LLM response.

Intervene or stop the agent

Click the Stop button at any time to interrupt the agent. The terminal returns to full manual control immediately. You can then continue typing manually or start a new agent instruction.

What you should see: The agent log shows "Stopped by user". The terminal is interactive again. Any partial changes the agent made remain on the screen.

Save the agent's actions as a script

After a successful agent run, click Save as Script in the agent panel. The agent's actions are converted to a JavaScript automation script and opened in the Script Editor.

What you should see: The Script Editor opens with the generated script. You can edit it, add loops or conditions, and save it for later replay via Tools > Macro Manager.

Best practices

Agent Mode works best when your instructions are specific. A few tips for reliable results:

Name the screen you expect: "On the Sign On screen, type…"
Provide exact values: "Enter customer number 00042"
Use F-key names the agent understands: "Press F6 to add a new record"
For local/offline use, install Ollama and select the Ollama provider, no API key needed
Keep instructions to one task at a time; chain tasks by running the agent multiple times

What you should see: The agent completes its task and shows a summary in the Agent Log. Detailed step logs let you audit exactly what the agent did and why.