@@ -24,21 +24,27 @@ Once enabled, there are a few ways to reach it.
 ### The quick prompt bar
 
 Press <kbd>`</kbd> (backtick, top-left of most keyboards) to pop up a prompt bar at the
-bottom of the viewport. Type a request, press <kbd>Enter</kbd>, and the assistant window
-opens so you can watch it work. Press <kbd>`</kbd> again or <kbd>Esc</kbd> to dismiss the
+bottom of the viewport. Type a request, press <kbd>Enter</kbd>, and the Chat window
+opens on its **Assistant** tab so you can watch it work. Press <kbd>`</kbd> again or <kbd>Esc</kbd> to dismiss the
 bar. It stays hidden until you summon it, so it never gets in the way.
 
-### The assistant window
+### The Assistant tab of the Chat window
 
-A floating, draggable window (it tabs and docks like Chat and the Explorer). It keeps the
-conversation, so you can refine across turns — "make them taller", "now paint the front
-one blue". Use **Stop** to cancel a run in progress.
+*Since 1.65* the assistant is not a window of its own: it is the **Assistant** tab of the one **Chat** window, beside
+**Session** (the people in this session). The window floats, docks in a sidebar or the bottom dock, and on a phone
+opens as one sheet. It keeps the conversation, so you can refine across turns — "make them taller", "now paint the
+front one blue". Use **Stop** to cancel a run in progress.
+
+The clock button opens **History**: your own provider's chats kept on this device and AI agent chats saved in the
+scene, next to the session chats. Your own provider's chats stay on this device (they are never sent to the session)
+and come back after a reload. See [Chat and the AI assistant](../menus-and-panels.md#chat-and-the-ai-assistant).
 
 ### The bottom-left button
 
 Once the assistant is enabled and a provider is set, a round **AI assistant** button (a sparkles icon) appears at the
-bottom-left of the viewport (just below the **＋** add button). Click it to open or close
-the assistant window — the same window the quick prompt bar opens into.
+bottom-left of the viewport (just below the **＋** add button). Click it to open the Chat window on its **Assistant**
+tab — the same tab the quick prompt bar opens into. If the window is on **Session**, the button switches to
+**Assistant**; if **Assistant** is already showing, it closes the window.
 
 ### In VR
 
@@ -67,10 +73,10 @@ The Voice typing settings, set to Groq.
 2. The **API key** for that provider; it stays on this device. Further down, **Test connection** sends half a second of silence.
 </div>
 
-The assistant window then has a **mic** button next to **Send**: click to listen, click again to stop. The words land in
+The **Assistant** tab then has a **mic** button next to **Send**: click to listen, click again to stop. The words land in
 the input, so you can fix them before sending. (In VR they are sent as soon as you let go.)
 
-![The AI assistant window: two of your messages and the assistant's answer, the provider picker, the input Ask the assistant…, the mic button and Send. 1: the mic button.](../img/ai-assistant/assistant-mic.webp){ .shot width="500" height="440" loading=lazy }
+![The AI assistant window: two of your messages and the assistant's answer, the provider picker, the input Ask the assistant…, the mic button and Send. 1: the mic button.](../img/ai-assistant/assistant-mic.webp){ .shot width="498" height="382" loading=lazy }
 /// caption
 The assistant window with the mic button next to Send.
 ///
 
@@ -14,7 +14,7 @@ replicated to everyone and undoable, exactly like a person's edits.
     (objects, flow nodes, physics, and mesh generation through *your* provider).
 
 !!! tip "New in 1.50: talk to it from the app"
-    Chat with the agent in the AI assistant window, send it pictures, and let it import models and files from its own
+    Chat with the agent in the Chat window's **Assistant** tab, send it pictures, and let it import models and files from its own
     machine — see [Talk to your agent](talk-to-your-agent.md).
 
 ## For you, the person in the session
@@ -136,7 +136,7 @@ before a join, scene tools answer *not connected — join first*. Then call `con
   back, the agent asks to come in again — the ordinary approval card — and carries on.
 - To stop it, revoke its invite or use **Remove from session** on its row ([Users and moderation](../people.md)).
 
-![The AI assistant window with Agent activity open: under the agent's name, Claude, two timed lines — created 3 objects (Rock1, Rock2 +1) and created 1 object (Flame) — each with Undo. 1: the number of steps so far. 2: Undo for one step.](../img/ai-build-with-an-agent/agent-activity.webp){ .shot .fade-b width="500" height="214" loading=lazy }
+![The AI assistant window with Agent activity open: under the agent's name, Claude, two timed lines — created 3 objects (Rock1, Rock2 +1) and created 1 object (Flame) — each with Undo. 1: the number of steps so far. 2: Undo for one step.](../img/ai-build-with-an-agent/agent-activity.webp){ .shot .fade-b width="498" height="214" loading=lazy }
 /// caption
 Agent activity in the AI assistant window.
 ///
 
@@ -33,15 +33,16 @@ like your own edits.
 ## Connect your tab (each time you build)
 
 1. Open theprototype.app, sign in, and open or join the scene you want to work on.
-2. Press the **AI** button and press **Connect Claude** (also in **Settings ▸ Connected apps**).
-   A chip reads **Claude connected · this tab**. The AI button always opens the window; signed out,
+2. Press the **AI** button and press **Connect Claude** (also in **Settings ▸ Account ▸ Connected apps**).
+   A chip reads **Claude connected · this tab**. The AI button always opens the Chat window on its **Assistant** tab
+   (since 1.65 the AI assistant is a tab of the one Chat window); signed out,
    the same card offers **Sign in to connect**, which opens your profile menu at the top right, the one place to
    sign in. With no AI provider set up, the window also offers **Use your own AI provider** (**Open settings**).
    Working with a local model or your own key while signed out? **Sign in to connect** stays in the window's header.
 3. Ask Claude something: *"Build a campfire with five logs and a warm light"*, *"Make the flame
    flicker"*, *"What can you do here?"*
 
-![The AI assistant window, signed out: Choose how the assistant thinks, with the Connect Claude card — Sign in to connect, step 1 Add theprototype to Claude with the connector address and How it works, step 2 sign in and press Connect Claude. 1: Sign in to connect. 2: Add to Claude. 3: How it works.](../img/ai-claude-connector/connect-card.webp){ .shot width="520" height="616" loading=lazy }
+![The AI assistant window, signed out: Choose how the assistant thinks, with the Connect Claude card — Sign in to connect, step 1 Add theprototype to Claude with the connector address and How it works, step 2 sign in and press Connect Claude. 1: Sign in to connect. 2: Add to Claude. 3: How it works.](../img/ai-claude-connector/connect-card.webp){ .shot width="518" height="574" loading=lazy }
 /// caption
 The AI window's Connect Claude card, before you sign in.
 ///
@@ -244,11 +245,12 @@ Claude reads its tool list when it connects, so it sees these after the connecto
 
 ## Stop it
 
-- **Disconnect** (next to the chip, or in Settings ▸ Connected apps) stops Claude at once, including
+- **Disconnect** (next to the chip, or in **Settings ▸ Account ▸ Connected apps**) stops Claude at once, including
   a step it is in the middle of. Press Connect Claude again to continue.
-- **Revoke** in **Settings ▸ Connected apps** removes the connector's access to your account
-  completely; claude.ai has to be approved again before it can work in any tab. You can also
-  remove the connector in claude.ai's settings.
+- **Revoke** in **Settings ▸ Account ▸ Connected apps** removes the connector's access to your account
+  completely; the AI app (claude.ai) has to be approved again before it can work in any tab. **Disconnect
+  everything** there does it for every connected app at once. You can also remove the connector in claude.ai's
+  settings.
 
 ## Privacy and limits
 
 
@@ -153,9 +153,10 @@ GLB. (Other hosted providers with the same task-based REST shape can be added as
 
 ## Using it
 
-- **Add menu → Generate 3D model…** — a small dialog: pick a provider, type a prompt,
-  optional name, **Generate**. The mesh lands where you opened the menu.
-- **From a picture** — the same dialog takes a photo, an Explorer picture or the viewport (Meshy, or a
+- **Add menu → Generate 3D model…** — the **Generate** window: pick a provider, type a prompt,
+  optional name, **Generate**. The mesh lands where you opened the menu. You can also open it from the dock's
+  **+** ▸ **Generate 3D**. See [The Generate window](#the-generate-window) below.
+- **From a picture** — the same window takes a photo, an Explorer picture or the viewport (Meshy, or a
   ComfyUI workflow with `{{IMAGE}}` in a LoadImage node). Meshy asks first (credits, and that the picture
   goes to Meshy). See [Give the AI pictures and shaders](pictures-and-shaders.md).
 - **AI assistant** — just ask (“add a treasure chest by the wall”). The assistant calls
@@ -165,6 +166,21 @@ GLB. (Other hosted providers with the same task-based REST shape can be added as
   (or `AGENT_MESH_*` env vars) and the agent's `generate_mesh` tool works the same way,
   pushing the GLB into the session. See [Console Agent](console-agent.md).
 
+### The Generate window
+
+*Since 1.65* the generator is a window, not a dialog. It stays open while you work, and you can dock it as a
+bottom-dock tab or a sidebar tab. On a phone it is a sheet. The form has not changed: provider, an optional picture
+(device, camera, viewport, Explorer), prompt, name. The credits line and the picture consent work as before.
+
+- **Jobs**: every generation on this device. A running job shows its progress and a **Cancel** button. A finished job
+  has **Reopen**, which selects the model, or places it again from Explorer ▸ Generated if you deleted it. A failed or
+  cancelled job has **Retry**, which is a new generation and uses credits again. **Clear finished** tidies the list.
+- **History**: the prompts you generated, newest first. Tap one to put it back in the form.
+- The jobs and the history survive a reload. A job that was still running carries on where it was: it is **not** sent
+  to the provider again, so it is not charged twice.
+- Pictures are never kept, so a picture job's **Retry** asks you to pick the picture again. Jobs and history stay on
+  this device (40 jobs, 30 prompts). Closing the window does not stop running jobs.
+
 ## Limits
 
 - Generated meshes replicate as full GLB data — very heavy meshes (>40 MB) are rejected, and
 
@@ -5,7 +5,7 @@ does not bundle or run it; your agent does, on your own machine and your own mod
 
 img2threejs rebuilds an object from **one picture** as procedural three.js code: a `THREE.Group` made of primitives and
 generated geometry, with a hierarchy of **pivots and sockets** (a lid that opens, a flame that flickers, a spot to hang a
-pot). With the AI assistant window and the `import_model` tool, the result lands in your scene as an ordinary model that
+pot). With the Chat window's **Assistant** tab and the `import_model` tool, the result lands in your scene as an ordinary model that
 every peer sees, saves with the scene and undoes in one step.
 
 ## 1. Install the skill beside the agent bridge
@@ -19,7 +19,7 @@ Needs Node.js ≥ 18, Git and Python ≥ 3.10. Restart the agent. Connect it to
 (`npx @theprototype/agent …`) and tell it to listen.
 
 ## 2. Send a picture in the chat
-In the AI assistant window pick your agent, then press **+** → *From device* / *Camera* (phones) / *From Explorer*, paste,
+In the Chat window's **Assistant** tab pick your agent, then press **+** → *From device* / *Camera* (phones) / *From Explorer*, paste,
 or drop the picture on the window. It is shrunk on your device (long edge 1568 px, about 1 MB) and its **location data
 is removed before it is sent**. The agent sees the picture, and the bridge also saves it to a temporary file whose path
 comes with your message — that path is what the skill takes.
 
@@ -21,7 +21,8 @@ Location data (EXIF, GPS) is removed from every picture **on your device, before
    pictures as it is. A ComfyUI workflow takes them when it has a `{{IMAGE}}` placeholder in a
    **LoadImage** node (see [3D Generation](generation.md)).
 2. **Add menu ▸ Generate 3D model…** and pick the picture: **From device…**, **Camera** (on a phone),
-   **Viewport**, or **From Explorer…**. You can also paste or drop a picture onto the dialog. In the
+   **Viewport**, or **From Explorer…**. You can also paste or drop a picture onto the window (a dockable
+   window since 1.65, see [The Generate window](generation.md#the-generate-window)). In the
    Explorer, a picture's menu has **Make a 3D model…**.
 3. A prompt is optional. With Meshy's *Refine* mode it becomes the texture prompt.
 4. Press **Generate**.
 
@@ -9,7 +9,7 @@ takes in between. The agent's terminal stays a second window onto the same conve
 
 ## Pick the agent
 
-Open the AI assistant window and choose the agent in the target list at the top, for example
+Open the Chat window's **Assistant** tab (the AI button) and choose the agent in the target list at the top, for example
 **Claude · agent via MCP** or **grok-3-mini · agent REPL**. Your configured assistants stay in the same list; a
 session can use both.
 
 
@@ -164,7 +164,7 @@ A file without these extras (a `.vrma` from another tool, an animation `.glb`) i
 
 ## Game-ready clips (Animation window ▸ clip tools)
 
-![The clip tools in the Animation window for the Knight's Walking_A clip: fps, frames and length with seamless loop, In place or Moving, loop, speed, Rename, Loop it and Remove, the playhead and in / out markers, Mark in, Mark out, Trim and Split at playhead. 1: the loop seam. 2: Loop it and Remove.](img/characters-and-motion/clip-tools.webp){ .shot width="372" height="443" loading=lazy }
+![The clip tools in the Animation window for the Knight's Walking_A clip: fps, frames and length with seamless loop, In place or Moving, loop, speed, Rename, Loop it and Remove, the playhead and in / out markers, Mark in, Mark out, Trim and Split at playhead. 1: the loop seam. 2: Loop it and Remove.](img/characters-and-motion/clip-tools.webp){ .shot width="370" height="443" loading=lazy }
 /// caption
 The clip tools in the Animation window.
 ///
 
@@ -240,7 +240,7 @@ dynamic.
 The other body travels with the trigger, so every peer reads the same object. A contact with the ground plane reports an
 empty *other*.
 
-![The node editor: an On Impact node whose only with input is wired from the Target's Object Selector and whose other output feeds a Distance node, and an On Enter node on a sensor zone filtered to the Ball. 1: only with. 2: other, the body that was hit.](img/colliders/contact-filter.webp){ .shot width="659" height="361" loading=lazy }
+![The node editor: an On Impact node whose only with input is wired from the Target's Object Selector and whose other output feeds a Distance node, and an On Enter node on a sensor zone filtered to the Ball. 1: only with. 2: other, the body that was hit.](img/colliders/contact-filter.webp){ .shot width="661" height="361" loading=lazy }
 /// caption
 On Impact and On Enter nodes with only with wired to Object Selectors.
 ///
 
@@ -229,7 +229,8 @@ The site and the app share one sign-in. **Sign in** in the site header offers th
 ### Sign-in methods and "Can't sign in?"
 
 *New in 1.54.1.* theprototype.app has no passwords: you sign in with GitHub or Google. If you lose access to that one
-account, you lose the way in. Link both now and either one gets you back: **Account menu ▸ Sign-in methods**.
+account, you lose the way in. Link both now and either one gets you back: **Settings ▸ Account ▸ Sign-in methods**
+(account menu ▸ **Manage**; it was in the account menu itself until 1.64).
 
 ![Account menu ▸ Sign-in methods](img/community/sign-in-methods.png)