vibe64 0.1.39 → 0.1.40

package/dist/index.html

@@ -21,72 +21,80 @@
       }, true);
     </script>
     <style>#startup-shell{color:#25232a;background:#faf9fc;grid-template-rows:56px 1fr;font:15px system-ui,sans-serif;display:grid;position:fixed;inset:0}#startup-shell header{border-bottom:1px solid #dedce3;align-items:center;padding:0 24px;font-weight:600;display:flex}#startup-shell main{grid-template-columns:240px 1fr;min-height:0;display:grid}#startup-shell aside{border-right:1px solid #dedce3;padding:24px}#startup-shell .startup-placeholder{background:#e5e2eb;border-radius:4px;height:16px;margin-bottom:20px}#startup-shell section{padding:32px}#startup-shell a{min-height:48px;color:inherit;align-items:center;display:inline-flex}@media (width<=600px){#startup-shell main{grid-template-columns:1fr}#startup-shell aside{display:none}}@media (prefers-color-scheme:dark){#startup-shell{color:#eeeaf3;background:#19181d}#startup-shell header,#startup-shell aside{border-color:#3c3942}#startup-shell .startup-placeholder{background:#35323d}}</style>
-    <script type="module" crossorigin src="/assets/index-dNxr4Lg9.js"></script>
+    <script type="module" crossorigin src="/assets/index-B9P6KjvH.js"></script>
+    <link rel="modulepreload" crossorigin href="/assets/rolldown-runtime-DBalRfVu.js">
     <link rel="modulepreload" crossorigin href="/assets/runtime-core.esm-bundler-DGXg_mpN.js">
-    <link rel="modulepreload" crossorigin href="/assets/ripple-BVltMmyV.js">
+    <link rel="modulepreload" crossorigin href="/assets/ripple-EFNA8EBu.js">
     <link rel="modulepreload" crossorigin href="/assets/normalize-CO2KHj_d.js">
     <link rel="modulepreload" crossorigin href="/assets/error-Dug56can.js">
     <link rel="modulepreload" crossorigin href="/assets/deepFreeze-BgoYkoTa.js">
-    <link rel="modulepreload" crossorigin href="/assets/useRealtimeEvent-Bwapstla.js">
-    <link rel="modulepreload" crossorigin href="/assets/useEndpointResource-DWoQ2keu.js">
+    <link rel="modulepreload" crossorigin href="/assets/vibe64ProjectScope.local-B06ABkhI.js">
+    <link rel="modulepreload" crossorigin href="/assets/useRealtimeEvent-DrOiU8lJ.js">
+    <link rel="modulepreload" crossorigin href="/assets/useEndpointResource-Ds8EDBZ2.js">
     <link rel="modulepreload" crossorigin href="/assets/useApi-BPuI6ZR9-CFWU1m37.js">
     <link rel="modulepreload" crossorigin href="/assets/surface-DY1u73tB.js">
     <link rel="modulepreload" crossorigin href="/assets/placement-CwTqWLQH.js">
     <link rel="modulepreload" crossorigin href="/assets/preload-helper-BvdcsxN9.js">
-    <link rel="modulepreload" crossorigin href="/assets/VAvatar-DPzqDH4r.js">
-    <link rel="modulepreload" crossorigin href="/assets/lazy-NHSwD9FS.js">
-    <link rel="modulepreload" crossorigin href="/assets/VOverlay-CeMZurIN.js">
-    <link rel="modulepreload" crossorigin href="/assets/createSimpleFunctional-CZTGuNuR.js">
+    <link rel="modulepreload" crossorigin href="/assets/lazy-BjyT2XvW.js">
+    <link rel="modulepreload" crossorigin href="/assets/VOverlay-Cp4rFk6g.js">
+    <link rel="modulepreload" crossorigin href="/assets/createSimpleFunctional-zlXZooQW.js">
+    <link rel="modulepreload" crossorigin href="/assets/dialog-transition-D2vqGksQ.js">
+    <link rel="modulepreload" crossorigin href="/assets/VAvatar-DvRBcmqv.js">
     <link rel="modulepreload" crossorigin href="/assets/ssrBoot-Rlh6kLbD.js">
-    <link rel="modulepreload" crossorigin href="/assets/VDivider-BMEqUoAB.js">
-    <link rel="modulepreload" crossorigin href="/assets/VMenu-C_x18lwb.js">
-    <link rel="modulepreload" crossorigin href="/assets/VChip-CcI-Oo2J.js">
-    <link rel="modulepreload" crossorigin href="/assets/VTextField-BR2eZeFT.js">
-    <link rel="modulepreload" crossorigin href="/assets/VSheet-DZWsu0kL.js">
-    <link rel="modulepreload" crossorigin href="/assets/VSelect-DL3HaA5m.js">
+    <link rel="modulepreload" crossorigin href="/assets/VDivider-Dq25WsKu.js">
+    <link rel="modulepreload" crossorigin href="/assets/VMenu-CDcBLYWR.js">
+    <link rel="modulepreload" crossorigin href="/assets/VChip-VVdtjfq-.js">
+    <link rel="modulepreload" crossorigin href="/assets/VTextField-C0b_0cSa.js">
+    <link rel="modulepreload" crossorigin href="/assets/VCheckboxBtn-DYcQU2Pn.js">
+    <link rel="modulepreload" crossorigin href="/assets/VSheet-DxYEFfnM.js">
+    <link rel="modulepreload" crossorigin href="/assets/VSelect-D3oLvCcB.js">
+    <link rel="modulepreload" crossorigin href="/assets/layout-B2Ap07aa.js">
     <link rel="modulepreload" crossorigin href="/assets/vibe64AssistantHost-CP6oxv5b.js">
-    <link rel="modulepreload" crossorigin href="/assets/useVibe64Learning-DyqXkpLw.js">
-    <link rel="modulepreload" crossorigin href="/assets/studioUrls-DmMMIScP.js">
-    <link rel="modulepreload" crossorigin href="/assets/mdi-CG9pWOSj.js">
-    <link rel="modulepreload" crossorigin href="/assets/VBtn-B2MBK2DL.js">
-    <link rel="modulepreload" crossorigin href="/assets/VAlert-P256xaE0.js">
-    <link rel="modulepreload" crossorigin href="/assets/VAutocomplete-BDb95qJX.js">
-    <link rel="modulepreload" crossorigin href="/assets/VCheckbox-C7A1ZW4D.js">
-    <link rel="modulepreload" crossorigin href="/assets/VDialog-Bu3EfKb1.js">
-    <link rel="modulepreload" crossorigin href="/assets/VCard-mXy0k4aS.js">
-    <link rel="modulepreload" crossorigin href="/assets/VRow-Cpw7OHQV.js">
-    <link rel="modulepreload" crossorigin href="/assets/VSpacer-Dhmsvk4t.js">
-    <link rel="modulepreload" crossorigin href="/assets/VExpansionPanels-Dgjrop0V.js">
-    <link rel="modulepreload" crossorigin href="/assets/VSkeletonLoader-CttxM1tw.js">
-    <link rel="modulepreload" crossorigin href="/assets/VSwitch-BRHeDJkH.js">
+    <link rel="modulepreload" crossorigin href="/assets/studioUrls-DpSS-0z1.js">
+    <link rel="modulepreload" crossorigin href="/assets/mdi-Bt90Ns76.js">
+    <link rel="modulepreload" crossorigin href="/assets/VBtn-BhVEwXLX.js">
+    <link rel="modulepreload" crossorigin href="/assets/VAlert-gwuhYlYb.js">
+    <link rel="modulepreload" crossorigin href="/assets/VAutocomplete-D7Vk5wq_.js">
+    <link rel="modulepreload" crossorigin href="/assets/VCheckbox-DyCGSN0H.js">
+    <link rel="modulepreload" crossorigin href="/assets/VDialog-CQDhjRiV.js">
+    <link rel="modulepreload" crossorigin href="/assets/VCard-C3Hx769y.js">
+    <link rel="modulepreload" crossorigin href="/assets/VRow-BskP28jR.js">
+    <link rel="modulepreload" crossorigin href="/assets/VSpacer-CgWOWvBh.js">
+    <link rel="modulepreload" crossorigin href="/assets/VExpansionPanels-C1y-myl5.js">
+    <link rel="modulepreload" crossorigin href="/assets/VRadioGroup-CAjNZwF7.js">
+    <link rel="modulepreload" crossorigin href="/assets/VSkeletonLoader-Cync3lZZ.js">
     <link rel="modulepreload" crossorigin href="/assets/_plugin-vue_export-helper-DNW4j-dv.js">
-    <link rel="modulepreload" crossorigin href="/assets/StudioErrorNotice-CSrUw1QV.js">
-    <link rel="modulepreload" crossorigin href="/assets/browserLocalStorage-B4C7rHrs.js">
     <link rel="modulepreload" crossorigin href="/assets/clipboard-CZs9aH3L.js">
-    <link rel="modulepreload" crossorigin href="/assets/useVibe64Terminal-Cg3idMcy.js">
-    <link rel="modulepreload" crossorigin href="/assets/shared-DpMvYJO3.js">
-    <link rel="modulepreload" crossorigin href="/assets/assistantRouting-D4XpXcEf.js">
+    <link rel="modulepreload" crossorigin href="/assets/Vibe64CopyCommand-CYiZcy0d.js">
+    <link rel="modulepreload" crossorigin href="/assets/VSwitch-BxGuGypi.js">
+    <link rel="modulepreload" crossorigin href="/assets/StudioErrorNotice-C_jbvp9V.js">
+    <link rel="modulepreload" crossorigin href="/assets/browserLocalStorage-B4C7rHrs.js">
+    <link rel="modulepreload" crossorigin href="/assets/useVibe64Terminal-I7Sf6xMW.js">
+    <link rel="modulepreload" crossorigin href="/assets/shared-DZETtWRH.js">
+    <link rel="modulepreload" crossorigin href="/assets/assistantRouting-DeE8jvtE.js">
     <link rel="modulepreload" crossorigin href="/assets/pathname-Bb895US8.js">
     <link rel="modulepreload" crossorigin href="/assets/usePaths-Db88-HOm.js">
-    <link rel="modulepreload" crossorigin href="/assets/useUiFeedback-BPBF_0bp.js">
-    <link rel="modulepreload" crossorigin href="/assets/useCommand-Bqxb-naz.js">
+    <link rel="modulepreload" crossorigin href="/assets/useUiFeedback-CMwE4MLs.js">
+    <link rel="modulepreload" crossorigin href="/assets/useCommand-nWFczmJy.js">
     <link rel="modulepreload" crossorigin href="/assets/shared-BEwrtWSp.js">
     <link rel="modulepreload" crossorigin href="/assets/vibe64ApiResponses-DNLY1Kel.js">
-    <link rel="modulepreload" crossorigin href="/assets/client-BWV84wi2.js">
-    <link rel="modulepreload" crossorigin href="/assets/VTextarea-DbqamXJd.js">
+    <link rel="modulepreload" crossorigin href="/assets/studioGateApi-BXHVBVEc.js">
+    <link rel="modulepreload" crossorigin href="/assets/client-CybfBKCn.js">
+    <link rel="modulepreload" crossorigin href="/assets/VTextarea-B7sZd1JY.js">
     <link rel="modulepreload" crossorigin href="/assets/thinkingText-DewCOEcu.js">
-    <link rel="modulepreload" crossorigin href="/assets/conversation-B90ERX9h.js">
-    <link rel="modulepreload" crossorigin href="/assets/useVibe64ProjectScope-Bp1ImZlz.js">
+    <link rel="modulepreload" crossorigin href="/assets/conversation-DZW4-Wck.js">
+    <link rel="modulepreload" crossorigin href="/assets/useVibe64ProjectScope-B5ogg6jd.js">
     <link rel="modulepreload" crossorigin href="/assets/vibe64BrowserTabOrigin-SF9E31vi.js">
-    <link rel="modulepreload" crossorigin href="/assets/useVibe64DatabaseTools-4o6S8eXq.js">
-    <link rel="modulepreload" crossorigin href="/assets/useVibe64SystemGraph-B4RmsC9Y.js">
+    <link rel="modulepreload" crossorigin href="/assets/useVibe64DatabaseTools-Dpm53ruL.js">
+    <link rel="modulepreload" crossorigin href="/assets/useVibe64SystemGraph-DbSfrYF8.js">
     <link rel="stylesheet" crossorigin href="/assets/ripple-DqhAt1Ze.css">
-    <link rel="stylesheet" crossorigin href="/assets/VAvatar-9ilDR2r2.css">
     <link rel="stylesheet" crossorigin href="/assets/VOverlay-DG-PukqI.css">
+    <link rel="stylesheet" crossorigin href="/assets/VAvatar-9ilDR2r2.css">
     <link rel="stylesheet" crossorigin href="/assets/VDivider-DvkP_iX7.css">
     <link rel="stylesheet" crossorigin href="/assets/VMenu-D_W8zRGm.css">
     <link rel="stylesheet" crossorigin href="/assets/VChip-PprDbmQn.css">
-    <link rel="stylesheet" crossorigin href="/assets/VTextField-BiatYyd3.css">
+    <link rel="stylesheet" crossorigin href="/assets/VTextField-F6unyhrS.css">
+    <link rel="stylesheet" crossorigin href="/assets/VCheckboxBtn-DCyuVH7W.css">
     <link rel="stylesheet" crossorigin href="/assets/VSheet-BbLSwuzR.css">
     <link rel="stylesheet" crossorigin href="/assets/VSelect-y7fezdfU.css">
     <link rel="stylesheet" crossorigin href="/assets/VBtn-uSpaN2Ha.css">
@@ -98,14 +106,16 @@
     <link rel="stylesheet" crossorigin href="/assets/VRow-DIr3-SqU.css">
     <link rel="stylesheet" crossorigin href="/assets/VSpacer-DfbUir7X.css">
     <link rel="stylesheet" crossorigin href="/assets/VExpansionPanels-cvCWEkWz.css">
+    <link rel="stylesheet" crossorigin href="/assets/VRadioGroup-C9f4tbih.css">
     <link rel="stylesheet" crossorigin href="/assets/VSkeletonLoader-LCPSxykp.css">
+    <link rel="stylesheet" crossorigin href="/assets/Vibe64CopyCommand-ClbQLYZi.css">
     <link rel="stylesheet" crossorigin href="/assets/VSwitch-FGoTnWb6.css">
     <link rel="stylesheet" crossorigin href="/assets/StudioErrorNotice-1K8E_MXD.css">
     <link rel="stylesheet" crossorigin href="/assets/useVibe64Terminal-Bk2QOSDO.css">
-    <link rel="stylesheet" crossorigin href="/assets/client-DGrP-2AJ.css">
+    <link rel="stylesheet" crossorigin href="/assets/client-D6UGamrv.css">
     <link rel="stylesheet" crossorigin href="/assets/VTextarea-DDl63AeD.css">
-    <link rel="stylesheet" crossorigin href="/assets/conversation-vxJGwmtW.css">
-    <link rel="stylesheet" crossorigin href="/assets/index-irlMYzLh.css">
+    <link rel="stylesheet" crossorigin href="/assets/conversation-TK6EmDk4.css">
+    <link rel="stylesheet" crossorigin href="/assets/index-BC29JK2e.css">
   </head>
   <body>
     <div id="app">

package/docs/colleague-usage/account-profile-and-appearance.md

@@ -23,7 +23,7 @@ name** stays disabled.
 
 Under **Appearance** (This browser), choose **Light**, **Dark** or **System**.
 The whole app switches immediately, including the Learning environment (**L**),
-its lesson reader and the Learn page's Lesson tab. **System** follows your device's light or dark setting
+its lesson reader and the Learn page's Lesson pane. **System** follows your device's light or dark setting
 and changes when the device does. Light is the default.
 
 The choice is saved in this browser only. Another browser, another device or a

package/docs/colleague-usage/{ai-accounts-and-routing.md → ai-accounts.md}

@@ -1,4 +1,4 @@
-# Connect AI accounts and choose Model routing
+# Connect AI accounts
 
 ## Try an app while verification continues
 
@@ -29,13 +29,14 @@ coding session is required. AI Accounts is the same screen as
 [AI onboarding](ai-onboarding.md): while no AI is connected (accounts marked
 **Reconnect required** do not count) it shows **Pick the AI behind Vibe64** with
 the ranked provider cards; once one is, it shows **Your AI setup** with your
-cards first, each with **Manage**, and the collapsed **Advanced** panel below
-them. Finishing a connection there returns to that overview (**Done**). The
+cards first, each with **Manage**, a **Models** summary and the collapsed
+**Advanced** panel below them. Which model does which job is explained in
+[AI model setups](ai-model-setups.md). Finishing a connection there returns to that overview (**Done**). The
 guided flow with its **Why** screen opens from **Redo onboarding** in the account
 menu (next to **Account settings**). Vibe64 includes no AI of its own; the free
 option is GLM-4.7 Flash through a free Z.AI account (see below). OpenCode Zen is
 an ordinary keyed provider: add your own Zen key to use it. Its free Big Pickle
-model is not offered, and model routing or Colleague choices that named it were
+model is not offered, and model or Colleague choices that named it were
 cleared by the release upgrade, so the next use recommends one of your connected
 AIs.
 
@@ -49,10 +50,10 @@ logins in your home folder, so a login made in a terminal works in Vibe64 too).
 opencode is different: Vibe64 runs your installed opencode program but signs it in with
 the keys you add in AI Accounts, not with opencode's own login.
 
-Before you pick a provider, its card in AI onboarding and **Your AI setup** says
-**Needs Claude Code on this computer** (or Codex, or opencode) under its buttons
-when that program is missing; the GLM free option says **Needs opencode on this
-computer**. When a program is missing, AI Accounts (and the matching step of AI onboarding) shows
+Before you pick a provider, a card in AI onboarding and **Your AI setup** whose program is missing is
+greyed and leads with **Install Claude Code first** (or Codex, or opencode), which
+opens the install row inside the card; a card whose program runs says **Ready on
+this computer** (see [AI onboarding](ai-onboarding.md)). When a program is missing, AI Accounts (and the matching step of AI onboarding) shows
 a warning instead of the sign-in button. In **Advanced → Add a connection**, an orchestrator whose
 program is missing says **Not installed on this computer** under its name; choosing it
 (or **Choose provider** for OpenCode) opens the same warning, and the setup dialog's
@@ -63,15 +64,14 @@ status chip reads **Not installed** or **Didn't start** instead of **Not connect
   **Other ways to install** it. Open a terminal on this computer, paste the command,
   press Enter, then choose **Check again**. You run the command yourself; Colleague
   can explain it but cannot run it for you.
-  The command matches the computer Vibe64 runs on. On macOS and Linux:
+  The same commands work on macOS, Linux and Windows inside WSL (Vibe64 runs on
+  Windows only inside WSL, so install them in the Ubuntu terminal, not PowerShell):
   - Claude Code: `curl -fsSL https://claude.ai/install.sh | bash`
   - Codex: `curl -fsSL https://chatgpt.com/codex/install.sh | sh`
   - opencode: `curl -fsSL https://opencode.ai/install | bash`
 
-  On Windows, in PowerShell (opencode uses its npm package, which needs Node.js):
-  - Claude Code: `irm https://claude.ai/install.ps1 | iex`
-  - Codex: `powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"`
-  - opencode: `npm install -g opencode-ai`
+  The setup guides walk through the whole computer: [Linux](set-up-linux.md),
+  [macOS](set-up-macos.md), [Windows](set-up-windows.md).
 - **Codex didn't start** means the program is installed but could not be run at all.
   Choose **Check again**; if it keeps failing, reinstall it with the same command. The
   program's own error output is under **Details**, collapsed.
@@ -85,24 +85,23 @@ needed. It looks on your `PATH` and in the official installers' folders
 (`~/.local/bin`, `~/.claude/local`, `~/.opencode/bin`, `/opt/homebrew/bin` and
 `/usr/local/bin`), so it finds them even when it was started outside a terminal. A
 program installed somewhere else must be on the `PATH` of the terminal that starts
-Vibe64. When Vibe64 starts, it prints one line for each program it can't find, and
-**Dashboard → Health** shows a check for each. Where the host provides pinned versions
+Vibe64. When Vibe64 starts with no AI program at all, it prints one plain message;
+**Dashboard → Health** needs one AI program and lists the others as optional. Where the host provides pinned versions
 of these programs, none of these warnings or install commands appear.
 
 ## Advanced in Your AI setup
 
 Below the provider cards, AI Accounts (and **Your AI setup** in AI onboarding)
-has a collapsed **Advanced** panel once any AI is configured. Its summary counts
-the connections, for example **7 connections · 2 not on a card above**. Select it
-to open:
+has a collapsed **Advanced** panel once any AI is configured. It holds only the connection
+plumbing; models are chosen in **Models** (see
+[AI model setups](ai-model-setups.md)). Its summary counts the connections, for
+example **7 connections · 2 not on a card above**. Select it to open:
 
 - **Add a connection** opens **Choose an orchestrator** (Claude Code, Codex, or
   OpenCode's provider catalog) for an exact orchestrator and provider pair. **Refresh** rechecks every account.
-- **Model routing** has one button per connected orchestrator (**Claude Code**,
-  **Codex**, **OpenCode**); each opens that orchestrator's routing form.
 - **Other connections** lists connected providers that have no card, such as
   OpenCode Zen or another OpenCode catalog provider, each **through OpenCode**,
-  with **Replace key**, **Model routing** and **Remove**. OpenCode Zen's model
+  with **Replace key** and **Remove**. OpenCode Zen's model
   access controls (**Check & enable working models**, **Enable all current
   models**, **Use the recommended model only**, **Cancel check**) are on its tile.
   These providers also appear in the **Connected** or **Currently using** strip
@@ -125,8 +124,9 @@ authorization code. You perform provider consent and secret entry. Wait for the
 saved connected state, and use **Refresh** or **Check saved key** when offered.
 Do not reconnect or resend a key solely because a status request was interrupted.
 DeepSeek, regular GLM API and GLM Coding Plan connections belong to the
-orchestrator shown in that row. Use **Advanced → Add a connection**, choose **Claude Code**,
-**Codex** or **OpenCode**, then its provider. Enter a key and complete
+orchestrator shown in that row. The DeepSeek card connects DeepSeek inside
+**Claude Code**; to connect it in another tool as well, use **Advanced → Add a
+connection**, choose **Claude Code**, **Codex** or **OpenCode**, then its provider. Enter a key and complete
 that row's verification. Each orchestrator can save a different key, or you can
 enter the same key independently for several orchestrators. Connecting or
 checking one row does not connect, check or copy a key to another. Regular GLM
@@ -151,16 +151,11 @@ known terminals using another provider keep running. Refusal preserves the saved
 key and connection status.
 OpenCode may require you to close its native terminals, other-provider chats and
 Colleague conversations before changing a key. If it refuses, the saved key stays
-unchanged and those conversations keep running; close the named users and retry. Saved routing can still name a disconnected
-provider; reconnect the exact row or deliberately choose a different route.
-**Start an AI session** lists only orchestrators you can use: one whose account
-was removed disappears until you connect it again, while its saved routing is
-kept. A connected orchestrator with a routing problem stays visible but greyed
-out with its reason; fix it with **Configure model routing**. An existing chat
-whose orchestrator is no longer connected shows **Open AI Accounts** and, when
-another orchestrator is available, **Switch to** that orchestrator (for example
-**Switch to Claude Code**). Switching keeps the chat's Senior, Junior or Auto
-mode (Custom becomes Senior), turns automatic review off, and keeps your draft.
+unchanged and those conversations keep running; close the named users and retry.
+Removing a connection keeps the saved model choices that name it; reconnect the
+exact row or deliberately choose other models in **Models**. Main chat always
+runs in your Senior's tool, so a removed Senior connection stops main chat until
+you reconnect it or choose another Senior.
 The controls and results are the same on desktop and mobile. Colleague can
 inspect redacted readiness and explain these steps; secret entry and connection
 consent remain your interaction in **AI Accounts**.
@@ -173,6 +168,8 @@ minute before retrying. If the error persists, ask the workspace operator to
 check the managed Claude version and refresh-lock recovery. This error alone
 does not prove that you signed out or need to replace your credentials.
 Changing the workspace's Claude account stops its existing Claude processes.
+Sign-ins and key checks show a spinner and what is happening while they wait,
+as described in [AI onboarding](ai-onboarding.md#while-it-works).
 Main chats keep their history and resume with the newly connected account; they
 do not require access to the previous account. Model selection in **Chat mode**
 also uses the current workspace connection. If the old process cannot be stopped,
@@ -199,83 +196,6 @@ change models or resend the pending words until cleanup and delivery are resolve
 Colleague can explain the error and inspect readiness, but cannot repair the
 host's process service. These recovery steps apply on desktop and mobile.
 
-After connecting, the **[provider] connected** screen names the orchestrator you
-selected. For example, connecting DeepSeek through Codex shows only Codex's
-optional suggestions and **Configure Codex routing**, even if the key also works
-with Claude Code. Review **Current** and **Suggested**, select the changes you
-want, then use **Apply [number] changes**. **Keep current routing** leaves the
-connection ready without applying those suggestions. If no changes are suggested,
-use **Done** or the same orchestrator's routing button under **Advanced → Model
-routing**.
-
-Open **Advanced → Model routing** and choose the orchestrator (or **Model
-routing** on a connection's tile) to choose its supported
-**Senior**, **Junior**, **Helper** and **Router** destinations and thinking choices.
-The displayed routing form can also offer **Review recommendations**,
-**Confirm Helper** and **Save routing**. Follow any
-required review, then Save routing. On mobile, scroll the same dialog to its controls.
-Disabled models are unavailable or incompatible; they must not be invented.
-
-The workflow picker checks access to your saved Senior, Junior and shared backup
-choices. It offers recommendations for roles that have not been configured;
-opening the picker does not replace saved choices or deliberately empty roles.
-First-time setup loads models for the chosen workflow and any other orchestrator
-already assigned to one of its roles. Choosing Codex or Claude Code does not
-require OpenCode model discovery unless one of those saved roles uses OpenCode.
-
-Recommendations prefer **Claude → Codex/OpenAI → DeepSeek → GLM** among models
-connected and permitted for each role. Within the chosen provider, Senior uses
-Opus/Astra, Junior uses Sonnet/Sol, and Helper/Router use Haiku/Luna when offered.
-Senior and Junior stay in the selected orchestrator; independent roles can use
-other permitted orchestrators. Personal connections cannot supply a shared
-backup. **No recommended changes** means the current choices already match the
-available recommendations. To review different recommendations for saved
-routes, select **Review recommendations**, inspect them, select **Apply to form**,
-then **Save routing**. Connecting or reconnecting does not overwrite saved roles.
-
-The model dropdown groups choices under orchestrator headings such as **Codex**,
-**Claude Code** and **OpenCode**. Choose an indented model beneath its heading;
-the second line identifies its provider and personal or workspace access.
-**Recommended** marks the suggested model for that role. Unavailable choices
-show their reason and cannot be selected. Type to search by model, orchestrator,
-provider or access; only matching groups remain visible. The selected field
-shows the model and its orchestrator/access details, without a thinking level.
-Use the separate **[role] thinking** control beside it to inspect or change
-thinking after selecting a model; the chosen level is included in Save routing.
-Recommendation reviews and collaborator outcomes still name the complete model
-and thinking selection. On mobile, the controls may be stacked vertically.
-
-Senior and Junior share an orchestrator. Helper and Router have their own allowed
-choices. Helper also reviews your work when you press **Check my work** on a
-lesson's review step: it uses the Helper of the practice session's workflow, on
-your own account. Without an available Helper the review says so and records
-nothing; choose one here, then press **Check my work** again. Shared Backup can supply a permitted alternative for collaborators.
-Changing routing applies to future work, not a running agent or Colleague's own
-model. Select Colleague's model through its own composer model picker; active work
-can prevent changing it until its turn stops.
-
-Colleague needs no Model routing step of its own. If an AI shows **Connected**
-but its roles were never saved (for example a Codex login that existed before
-Vibe64 was installed), Colleague's first message or **Teach me** saves that
-workflow's recommended routing, as starting a session does, and uses its
-Senior model, trying Claude Code before Codex and then the other connected
-workflows. Saved choices are never replaced. Colleague says it has **no usable
-AI model** only when no connected workflow can supply a Senior model; connect an
-AI, or open **Advanced → Model routing** and choose an available Senior, then press
-Send or **Teach me** again. That message clears by itself once an AI is usable.
-
-If the form says routing changed elsewhere, use **Reload current choices** and
-review your retained draft before saving. If no permitted model is usable, finish
-the real account/access setup before promising an AI answer. Vibe64 includes no
-AI of its own on any installation; AI Accounts or AI onboarding connects one,
-including the free GLM-4.7 Flash option.
-
-Colleague can explain these steps, inspect account/routing readiness, open AI
-Accounts and save supported routing changes when the owner explicitly asks. It
-offers assistance without changing anything for a how-to question. Connection,
-consent and model-access operations that are not exposed to it must be completed
-in the actual account UI; it should recheck status after the person finishes.
-
 For regular Z.AI setup, ask Colleague “Is my regular Z.AI connection saved?”
 The workspace owner can get a fresh, read-only check without opening a project.
 It reports the saved connection, preferred provider, default model and whether
@@ -295,9 +215,9 @@ If it reports insufficient balance or no resource package, check the provider's
 billing and quota before retrying. A failed replacement keeps the existing saved
 key. These messages and steps apply on desktop and mobile. Colleague can explain
 the reason; key entry remains your step in **AI Accounts**.
-The preferred provider/default-model badge does not prove future Senior routing
-or change Colleague's model. Ask for the actual routing before changing only the
-requested future workflow role. A correct lesson answer establishes understanding,
+The preferred provider/default-model badge does not prove which models Vibe64
+uses or change Colleague's model. Ask for the actual choices in **Models** before
+changing only the requested role. A correct lesson answer establishes understanding,
 not successful account setup or model execution.
 
 To connect the regular Z.AI API, use the GLM card's free option (**Start free**
@@ -324,9 +244,16 @@ If those choices changed while the dialog was open, press **Cancel**, reopen
 Claude Code uses its selected provider for background model calls as well.
 
 If an administrator's stopped-service upgrade reports an unsupported saved
-routing request even though the conversation uses current Senior, Junior, Auto
+model request even though the conversation uses current Senior, Junior, Auto
 or Custom, use the corrected candidate release's read-only upgrade preflight.
 Valid current choices and pending delivery receipts are preserved by the explicit
 upgrade. Do not reset the conversation or change its model to work around that
 error. Unknown or corrupt routing still needs administrator inspection before
 apply; Colleague can explain the recovery but cannot perform the host upgrade.
+
+Colleague can explain these steps, inspect account readiness and open AI
+Accounts. It offers help without changing anything for a how-to question.
+Connection, consent and model-access steps it cannot perform must be completed
+in the actual account screens; it rechecks status after you finish. Vibe64
+includes no AI of its own on any installation; AI Accounts or AI onboarding
+connects one, including the free GLM-4.7 Flash option.

package/docs/colleague-usage/ai-model-setups.md

@@ -0,0 +1,159 @@
+# Choose models: owner and team setups
+
+Vibe64 chooses a model for each job from **Models**. Open it from **Account
+settings → AI Accounts** (or AI onboarding's **Your AI setup**): the **Models**
+summary under the provider cards lists **Senior**, **Junior** and, in a
+workspace with a team, **Team members**; choose **Review models**. Connecting
+an AI is explained in [AI accounts](ai-accounts.md).
+
+Only the workspace owner changes models. Everyone else sees the team members'
+models read-only.
+
+## Jobs (roles)
+
+**Main chat**
+
+- **Senior**: Plans, reviews and Deslop in main chat. Your most capable model.
+- **Junior**: Everyday changes in main chat.
+- **Router**: Chooses Senior or Junior when main chat is on Auto.
+
+**Helpers**
+
+- **Colleague**: The default for Colleague. Each person can change their own.
+- **Fix it with AI**: Repair chats opened from errors and conflicts. Can run in a
+  different tool from the chat it fixes.
+- **Explain & check**: Explain this code, the database assistant and Check my
+  work.
+- **Quick helper**: Titles, commit messages, prompt hints and summaries.
+
+Main chat always runs in your Senior's tool (Claude Code, Codex or OpenCode);
+Senior and Junior must run in the same tool. A chat that last ran in another
+tool catches up on its next message. Helpers can each use any allowed model.
+
+## Two setups: Owner and Team members
+
+There are two setups. **Owner (you)** has every role and is used for main chat
+and for the owner's own helpers. **Team members** has only the helpers and is
+used by everyone else in the workspace for their Colleague, Fix it with AI,
+Explain & check and Quick helper. Main chat always runs on the owner's setup,
+and the owner approves what team members send to it.
+
+The switch **Owner (you) | Team members** at the top of Models appears only for
+the owner, and only when the team setup has a model or Models was opened for
+the team (from the team card or **Set up team AI**). A solo owner never sees it.
+
+On **Team members** the screen says: Team members' Colleague, Fix it with AI,
+Explain & check and Quick helper use these models. Main chat always runs on your
+setup, and you approve what team members send to it. Personal plans, like your
+Claude plan, can't be used by team members. Their model lists show only
+**Workspace use** models; saving a personal one is refused with **Team members
+can't use personal plans. Choose a Workspace-use model.**
+
+With nothing chosen yet it shows **No team AI yet. Team members can't open
+projects until you add one.** and **Add a DeepSeek key** (see
+[team AI access](team-ai-access.md)); from chat's Models it opens AI setup on
+the team step. When a Workspace-use model is already connected it adds **Or
+choose a Workspace-use model you already connected below.** with the Colleague
+and Quick helper pickers; Fix it with AI and Explain & check show their default
+**Same as** lines, saved with the first choice.
+
+## Same as
+
+A helper can follow another role. It shows one quiet line, for example **Same as
+Senior · Opus 5.5 · High**, with **Change**. By default the owner's Colleague
+follows Senior, Fix it with AI follows Junior and Explain & check follows Quick
+helper; the team's Fix it with AI follows the team's Colleague and Explain &
+check follows its Quick helper. A followed role changes with the role it
+follows.
+
+**Change** opens that row's model and thinking pickers with the followed
+choice; saving makes it the row's own choice. A row with its own choice offers
+**Use Senior's model** (or the role it would follow) to follow again.
+
+## Model, thinking and Runs in
+
+Each row shows its model and thinking, for example **Opus 5.5 · High**. Pick a
+model from the list: each item shows the model, its provider and an amber
+**Personal use** badge for a personal plan (see
+[personal-use subscriptions](ai-personal-use.md)). **Recommended** marks the
+suggested model; unavailable choices show their reason and cannot be chosen.
+Type to search. Choose thinking in the separate thinking control beside it.
+
+The tool a model runs in appears only when it is not the usual one, for example
+**DeepSeek V4 Pro · runs in Codex**. **Advanced** under a row appears only when
+there is something to see:
+
+- A subscription model runs in its own tool, with no choice: **Runs in Claude
+  Code — required by your Claude plan.** or **Runs in Codex — required by your
+  ChatGPT plan.**
+- A key model connected in more than one tool shows **Runs in** with one option
+  per connected tool (**Claude Code**, **Codex**, **OpenCode**) and
+  **Recommended** on the usual one: Claude Code for DeepSeek, Codex for an
+  OpenAI key and the GLM Coding Plan, OpenCode for the free GLM option, an
+  Anthropic key and other OpenCode providers.
+- When the provider's key is connected in only some tools, **Connect [provider]
+  in another tool** opens **Add a connection** (from chat's Models, AI setup
+  on that provider).
+
+Choosing another model keeps the row's tool when the new model runs there, and
+otherwise uses the new model's usual tool.
+
+## Recommendations and saving
+
+Roles nobody chose get recommended models when an AI is first connected,
+preferring Claude Code, then Codex/OpenAI, DeepSeek and GLM among models you may
+use. The owner's gold standard is Opus 5.5 · High for Senior, Opus 5.5 · Medium
+for Junior and Haiku 5.5 · Low for Router and Quick helper; the team's is
+DeepSeek V4 Pro · High for Colleague and DeepSeek V4.1 Flash · Low for Quick
+helper. Connecting or reconnecting never replaces saved choices.
+
+When recommendations differ from saved choices, **Review recommendations**
+lists the changes; **Apply to form** puts them in the form, then **Save**.
+**No recommended changes** means the choices already match. While Vibe64 checks
+the new choices it shows **Checking access…**. **Cancel** discards the draft.
+If Models changed in another tab, use **Reload current choices** and review
+your kept draft before saving. Changes apply to future work, not to a running
+agent.
+
+After connecting a provider, **[provider] connected** may suggest changes with
+**Current** and **Suggested**: select the ones you want and choose **Apply
+[number] changes**, or **Keep current models**. **Configure AI setup** opens
+Models.
+
+## Main AI
+
+**Main AI** is the provider behind your Senior. **Your AI setup** reads
+**Currently using · Main AI: [provider]**, adding the tool when it is not the
+provider's own, for example **Main AI: DeepSeek, via Claude Code**. **Use as
+main AI** on another card moves Senior, Junior, Router and Quick helper to that
+provider in the tool it is connected through (see
+[AI onboarding](ai-onboarding.md#redo-with-an-existing-setup)). The crown and
+**Recommended** ribbon stay on Claude's card; they never mean "in use".
+
+## Colleague's model
+
+Each person's Colleague starts on their setup's Colleague model: the owner's
+follows Senior, team members use the team's Colleague. Anyone can change their
+own Colleague's model from Colleague's model menu; that choice is theirs alone
+and does not change Models. Team members see the note **Change your own
+Colleague's model from Colleague's menu.** Colleague needs no setup step of its
+own: if the roles were never saved, its first message saves the recommended
+models. Colleague says it has **no usable AI model** only when none can serve
+it; connect an AI or choose an available model in Models, then send again.
+
+Explain & check also reviews **Check my work** on a lesson's review step on
+your own account; without an available model the review says so and records
+nothing.
+
+## Desktop and phone
+
+On a desktop each row has its label on the left and its pickers on the right. On
+a phone (under 600 px) rows stack label, description, model, thinking and
+Advanced, and Save/Cancel stay at the bottom.
+
+## What Colleague can do
+
+Colleague can explain the roles and setups, read the current choices, open AI
+Accounts and save supported model changes when the owner explicitly asks. It
+offers help without changing anything for a how-to question. Team members'
+Colleague can explain the team setup but cannot change it.