Agent Forge is StationOne’s workspace for creating and managing AI agents that automate multi-step workflows. Agents can use tools, knowledge bases, skills, and even other agents to accomplish complex tasks.
Creating an Agent
Option A: AI-Generated Agent (Recommended)
- Click Agent Forge in the sidebar.
- Click New Agent.
- In the Generator step, describe what your agent should do in natural language (e.g., “Create a research assistant that searches the web, saves findings to files, and generates visualizations”).
- Optionally select a generation model.
- Click Create my agent — wait for generation.
- Review the generated agent (name, instructions, workflow steps).
- Click Review my agent to continue to the editor.
- Customize as needed, then save.
Option B: Manual Creation
- Click Agent Forge → New Agent → Skip & Create Manually.
- Complete the wizard steps:
- Step 1 — General Information: Name, Role (User Initiated or Utility), Description, Instructions/Goal.
- Step 2 — Model Configuration: Select engine and model. A warning appears if the model doesn’t support agentic capabilities.
- Step 3 — Workflow Design (see below for details).
- Step 4 — Triggers (see Scheduling section).
Building Workflows
Each agent workflow consists of one or more steps executed in sequence (or non-linearly with flow control).
Step Components
Each step includes:
- Description: What this step does (shown in the step header).
- Prompt: Instructions for the AI. The prompt toolbar provides:
- Insert Variable: Insert previous step output, last run output, or create custom variables.
- Flow Control: Insert stop conditions or goto jumps (see Non-Linear Workflows).
- Memory: Write to memory or clear memory (see Agent Memory).
- Request Human Input toggle: When enabled, the agent pauses and waits for user approval before completing the step.
- Footer buttons: Attach a Skill, Knowledge Base, Tools, Co-Agents, or JSON Schema to this specific step.
Adding and Reordering Steps
- Click Add Step between existing steps to insert a new step.
- Drag and drop: Click and hold the grip handle (six dots) on the left side of any step to drag it to a new position. All variable references and goto targets are automatically reindexed.
- Click the trash icon in a step header to delete it.
Non-Linear Workflows
Stop Conditions
Insert a conditional exit point in a step’s prompt:
- Click Flow Control in the prompt toolbar → Insert Stop Condition.
- Enter a condition (e.g., “if the answer contains a definitive conclusion”).
- The agent stops the workflow early if the condition evaluates to true.
Goto (Jump to Step)
- Click Flow Control → Insert Goto.
- Specify the target step number.
- After the current step completes, the agent jumps to the specified step instead of proceeding sequentially.
- Use this to create loops or skip steps based on logic.
Agent Memory
Agents can persist information across multiple runs using memory.
Writing to Memory
In a step’s prompt toolbar, click Memory → Write to Memory. Describe what the agent should save.
Clearing Memory
Click Memory → Clear Memory to reset stored information.
Viewing and Editing Memory
From the Agent View screen, click the three-dot menu ( ⋮ ) → View Memory. A dialog shows the current memory content in an editable textarea. You can manually edit, save, or clear it.
NOTE: Memory is only available for multi-step agents and persists across runs within the same workspace.
Human-in-the-Loop
For steps that perform sensitive operations (file deletion, data updates, etc.):
- In the Workflow editor, toggle Request human input on for the relevant step.
- When the agent reaches that step during execution, it pauses.
- A confirmation request appears in the Inbox (and in chat if the agent was started from there).
- Review the action and click Approve (agent continues) or Reject (workflow stops).
Scheduling Agents
In the Triggers step of the agent editor:
- Select schedule type:
- At specific times (cron): Use the visual scheduler or enter a cron expression. Preset buttons for minutes, hours, days, weeks, months.
- At regular intervals: Set days, hours, and minutes between runs.
- Enable Catch-up to run missed executions when the app was closed:
- Options: Never, Within N minutes/hours/days, Always.
- Webhook Trigger (if enabled): Check Trigger this agent via HTTP request to get a webhook URL.
- Set default values for workflow variables used in scheduled/webhook runs.
Save Conversation as Agent
- In any chat, click the three-dot menu ( ⋮ ) → Save as Agent.
- Optionally provide guidance about the agent’s purpose.
- Click Create Agent.
- The system analyzes the conversation and generates an agent with appropriate workflow steps, tools, and model configuration.
- Review and customize in the full Agent Editor, then save.
Card View vs Table View
In the Agent Forge list, click the view toggle button (top-right):
- Card View: Grid of agent cards showing name, description, and action buttons.
- Table View: Compact list with columns for Name, Description, Type, Last Run, and Actions.
NOTE: The selected view is remembered across sessions.
Agent Management
Running an Agent
Click the Play button on an agent card/row. If the agent has variables, a dialog prompts for input values. Click Run to start execution.
Viewing Execution
The Agent View shows:
- Left pane: Agent info, action buttons (Run, Stop), and run history with status icons and timestamps.
- Right pane: Visual execution flow canvas showing step-by-step progress with status colors (running, success, pending, error). Click a step to see its details, prompt, response, and tool calls.
Import/Export
- Import: Click Import in the agent list → select a JSON file.
- Export: Right-click an agent → Export → saves as JSON.
- Duplicate: Right-click → Duplicate to create a copy.
Deleting
Right-click an agent → Delete → confirm in the dialog.