Agents Overview

Agents are saved, repeatable workflows that process your input through a fixed sequence of steps — combining AI, project context, and connected services to produce consistent output every time you run them.

Unlike free-form AI chat where you improvise turn by turn, an agent runs the same pipeline each time. You describe what you want once, approve the design, and then run it whenever you need it.

How agents differ from chat

Chat is open-ended. Each message is independent, the model improvises, and output stays in the conversation thread.
Agents are structured. They have a fixed step sequence decided at creation time. You attach input before running, the pipeline processes it, and results can be saved as files, displayed on screen, or sent to external services.

Agents are best for tasks you do repeatedly — summarizing documents, reformatting content, analyzing files, generating reports, or posting to external tools.

What agents can do

Depending on how they were designed, agents can:

Process content with AI — summarize, rewrite, analyze, combine, or transform documents
Display results — show output in the run interface with live streaming
Save files — write results to an output folder as Markdown, text, JSON, or other formats
Call connected APIs — post to GitHub, Slack, Notion, ClickUp, Linear, and other services
Search project memory — find relevant past content or conversations
Create or expand mind maps
Validate input before processing

Each agent only gets the capabilities that were included when it was created. It cannot invent new abilities at runtime.

How agents use project context

Before running an agent, you configure its context:

Input Sources — the files, folders, or workspace documents the agent will process
Reference files — optional context like brand guides, templates, or domain rules that inform how the agent works
Output Folder — where saved results go

The agent processes your input through its steps, with each step potentially building on the output of the previous one. This data flows automatically — you set up the context once and the pipeline handles the rest.

Where outputs are stored

Agent outputs appear in several places:

On screen during the run — each step shows its status and output in real time, with AI steps streaming text token by token
Saved files — if the agent includes a save step and you configured an output folder, results are written as files (named with the agent name and timestamp)
External services — API steps send data to the configured service
Execution log — each run is recorded internally for tracking

Where to find agents

Agents live on the Tools page in ThinkForge. From there you can create new agents, deploy existing ones, configure their context, and run them. You can also pin agents to the ribbon for quick access, run them from mind map nodes, or trigger them automatically on a schedule or when files change.