Chat

Chat is the home surface of AgentLoop and the primary way you get work done. You describe what you want in plain language, and AgentLoop figures out the rest — usually delegating to the Product Manager to break the work into tasks, or pulling in another agent for a specific request.

Talking to AgentLoop

Type a message and the chat agent routes your intent. You can manage tasks, dependencies, the board, and agent delegation entirely through conversation:

• “Break down user authentication into tasks and add them to the board.”
• “What’s the status of task 5?”
• “Have the engineer implement the login API endpoint.”
• “Ask QA to test the new authentication flow.”
• “Add a dependency: task 3 must finish before task 7 can start.”
• “Show me what’s blocked and why.”

Streaming and Live Activity

The assistant’s reply streams in token by token as it’s generated, so you start reading the moment the agent starts writing instead of waiting for the whole response to land.

While the agent works, an expandable “Thinking…” activity timeline shows live step and tool activity as it happens — every tool call and intermediate step, in real time. Expand it to follow along on a long turn, or leave it collapsed and just watch the answer arrive.

When the model exposes its own chain-of-thought, a collapsible “Reasoning” disclosure appears alongside the response. Expand it to see how the model arrived at its answer; it stays out of the way otherwise.

Changed your mind mid-answer? Hit Stop to cancel an in-flight response right away — the stream halts and you can refine your prompt and try again.

Inline Prompts

Some flows render right inside the conversation. When you open a new project, AgentLoop can run a guided spec workflow — an interactive questionnaire that appears inline so you can answer questions without leaving chat. Other inline elements (like behavior-tree subtrees) surface relevant detail in place.

Threads and History

Chat threads are persisted per project, so you can close the app and pick up exactly where you left off. The sidebar keeps a tree of your conversations, and you can run more than one thread per project.

Choosing a Model per Chat

Each chat session has its own model picker, so you can switch the model for a single conversation without touching your global default — reach for a faster model when you’re iterating, or a stronger one for tricky reasoning. If you land on a model you want everywhere, use the picker’s Set as default action to make it your global default without hand-editing any config.

Floating Overlay and Chat Dock

Chat doesn’t have to take over the window:

• Chat dock — a collapsible bar pinned to the bottom of every other view, so you can keep talking to AgentLoop without leaving the Tasks board or Agents workbench.
• Floating overlay — pop chat out into a frameless floating window that stays on top.

Asking About Your Codebase

Chat is preloaded with project context — specs, tasks, and a code index — so you can ask questions about the code itself and get answers grounded in your repository, not just generic responses.