Agent: overview

The built-in Flownix agent is a chat assistant that works inside your project: it plans, searches, writes specs and changes the tracker when you ask it to.

The agent is a chat assistant right in the Flownix web interface. It needs no Claude Code, no Cursor and no MCP token: open a project, click "Agent", and say what you need. The agent sees the work tree, documents, components, artifacts and decision history of the current project and acts in it on your behalf, with your permissions.

Agent panel next to a task — The agent panel open over a feature page: a /plan-feature request, collapsed reasoning, the action chain, “What was done” and the cost of the reply.

What it does

Plans

Breaks an idea down into an epic, features, plans and tasks using the tracker templates, and sets dependencies, tags and components.

Answers questions about the project

"What's in progress right now?", "Why did we choose Postgres?", "Which tasks touched billing?" — it searches nodes, documents, comments and artifacts by meaning and answers with links.

Maintains specs

Writes and edits documents, prepares spec deltas and submits them to you for approval. A person always approves.

Changes the tracker on request

Creates and moves nodes, changes statuses, writes comments, links tasks. Only when you ask — and deletion and other risky actions only after you click "Allow".

Works with files

Reads text files from Artifactory, saves its own reports there, and gives download links for everything else.

Shows what it did

Every tool call appears in the feed with its arguments and result, and each reply ends with a "What was done" summary linking to the changed entities.

What it is not

The built-in agent is a specialist in your Flownix project, not a general-purpose assistant and not a programmer.

It has no access to code. No repository, no file system, no command line. It can't write code, run tests or commit — for that you still connect an external agent over MCP and give it /execute-task.
It stays inside the project. Everything it reads and changes belongs to the current project. It can't create projects or organizations.
It doesn't make decisions for people. You approve spec deltas, you start harnesses, you confirm deletions.
It doesn't chat about anything else. The agent politely declines questions that are not about Flownix or your project.

Built-in agent or external agent over MCP

	Built-in agent	External agent (Claude Code, Cursor…) over MCP
Where it lives	In the Flownix web interface	In your editor or terminal
What you need to start	Nothing but access to the project	An MCP token, client setup, skills
Code and repository	No access	Full access to your working copy
How you pay	Agent credits from your Flownix plan	Your model provider subscription
Who sees the conversation	All project members	Only you, in your client
Best for	Planning, specs, questions about the project, tracker cleanup	Executing tasks in code, reviewing branches, agent councils

They don't compete; they work as a pair. The built-in agent helps you break a feature into tasks and write the spec; the external one picks up tasks by slug and implements them in code. Both write to the same tree, so no context gets lost.

Working in the chat — how to open the agent, manage sessions, invoke skills and answer its questions.
What the agent can do — the full list of actions, limitations and confirmations.
Credits and limits — what a turn costs and what your plan includes.
Use cases — what it looks like on real tasks.