---
id: F8
title: Chat
status: partial
order: 8
---

# F8 — Chat

## What it is

The **chat** is a first-class surface of the Tasks app — not a side panel bolted onto a traditional UI. Users talk to the **agent** in natural language; the agent performs the same operations the UI can, through the same **actions**. There is no separate agent workflow, permission model, or data path.

**Parity rule:** whatever a user can do in the UI, the agent can do in chat — and vice versa.

## What users can do

<!-- status: done -->

### Work through conversation

- **Capture** ideas ("remind me to call the plumber") — creates an **inbox item** by default; use **create-task** only when the user asks to add directly to the task list.
- **Manage tasks** — add, edit title, mark complete/incomplete, delete, reorder, bulk complete/delete ([F1](./f1-tasks.mdx)).
- **Triage inbox** — list, edit, delete, reorder, and **mark ready** ([F4](./f4-inbox.mdx)).
- **Filter** the task list the same way as the UI (e.g. hide completed tasks).
- **Ask what's on screen** — the agent can read the user's current view and selection before acting.
- **Keep the current page open while reviewing tasks** — when the user is elsewhere in the app, the agent can show an interactive task-list widget inline in chat.

<!-- status: todo -->

### Work through conversation (extended)

- **Assign** tasks to a [project](./f3-projects.mdx); open project pages.
- **Work with custom fields** — set Notes, Due date, Priority, and other fields; define new fields ([F2](./f2-custom-fields.mdx)).
- **Arrange lists** — filter, sort, group, and switch [saved views](./f5-views.mdx).
- **Find things** — [search](./f6-search.mdx) by task title or project name and open a result.

### Delegate from the UI

Buttons and shortcuts in the UI may **send a message to the chat** instead of duplicating logic — e.g. "Add with agent", "Triage inbox", "Summarize my open tasks". The user sees and steers the work in the thread; the agent calls the same actions the UI would.

## What the agent can do (parity map)

Every UI capability maps to agent tools (actions). The agent uses these — not ad-hoc scripts or direct database access.

<!-- status: done -->

| Area                                           | Agent can…                                                                                               |
| ---------------------------------------------- | -------------------------------------------------------------------------------------------------------- |
| **[F1 Tasks](./f1-tasks.mdx)**                 | Add, edit, delete tasks; mark complete/incomplete; reorder; bulk complete/delete; apply completed filter |
| **[F4 Inbox](./f4-inbox.mdx)**                 | Add, edit, delete inbox items; reorder; mark ready (promote to task)                                     |
| **[F2 Custom fields](./f2-custom-fields.mdx)** | Create, rename, configure options; set field values; delete fields                                       |
| **Navigation**                                 | Open inbox, main task page, or a specific task via `navigate`                                            |
| **Context**                                    | Read current screen and bulk `selection` via `view-screen`                                               |
| **Inline task list**                           | Render the task list in chat with `render-task-list-inline` when the user is not on `/tasks`             |

<!-- status: todo -->

| Area                                 | Agent can…                                                            |
| ------------------------------------ | --------------------------------------------------------------------- |
| **[F4 Inbox](./f4-inbox.mdx)**       | Bulk mark ready / delete; assign project (hidden until ready)         |
| **[F1 Tasks](./f1-tasks.mdx)**       | Bulk assign project                                                   |
| **[F3 Projects](./f3-projects.mdx)** | Create, rename, delete projects; list tasks in a project              |
| **[F5 Views](./f5-views.mdx)**       | Apply filter, sort, group; create, switch, rename, delete saved views |
| **[F6 Search](./f6-search.mdx)**     | Search tasks and projects; open a result                              |
| **Navigation**                       | Open inbox, project pages                                             |

If a capability is added to the UI, it is exposed as an action and documented for the agent in the same change.

## How it behaves

### Same rules as the UI

The agent follows every product rule that has shipped; extended rules apply as features land.

<!-- status: done -->

- New capture creates an **inbox item** by default; **create-task** only when the user asks to add directly to the task list.
- **Destructive** actions require **confirmation** — in chat, the agent asks before delete (same as UI confirmation dialogs).

<!-- status: todo -->

- **Completion** on inbox items is not allowed until **mark ready**.
- Inbox items with a project assigned do **not** appear on the project page until **mark ready**.
- **Bulk** actions require **confirmation**.
- **Search** matches titles and project names only — not custom field values.
- **Sort by field** disables manual reorder until sort is **None** again.

### Context-aware

- **`view-screen`** is the first call when the user's visible context matters — which page, selected tasks, active filter, and (when shipped) project, view, or search query.
- **`navigation`** application state stays in sync with the UI so the agent knows what the user is looking at without guessing.
- After the agent changes data, the **UI updates automatically** (same real-time sync as UI-initiated writes).

<!-- status: done -->

### Capture

Typical flow:

1. User says something in chat → **inbox item** created (title from the message) unless they ask for the task list directly.
2. User refines in chat → edit inbox item or mark ready to promote to a task; manage tasks on `/tasks` as usual.

<!-- status: todo -->

### Capture and triage (extended UI)

Full inbox page UI:

1. User captures on the inbox page or triages in chat.
2. User refines on the inbox or task page → edit title, fields, project, etc.
3. User says "mark it ready" → inbox item becomes an **incomplete** task.

### No shadow AI

The UI does not call an LLM directly for product work. Generative or interpretive steps go through the **agent chat** so the user can see, correct, and audit what happened.

## Limits (v1)

- One agent per app instance — no multi-user agent handoff.
- Chat does not add capabilities beyond what the UI has shipped; it is the conversational interface to the same actions.
- No voice-only or external messaging channels (Slack, email) in v1.