refactor(chat): drop the unreachable 'mixed' content type, widen messageContent, tokenize the popup launcher

Three small cleanups:

- ContentType no longer carries 'mixed'. createContentClassifier never emitted
  it, so the `|| currentType === 'mixed'` branch in the delta path was dead and
  any consumer switch on it had a case no input reaches. The docs said as much
  and told readers not to branch on it; the member is simply gone instead.

- messageContent() takes { content: unknown } rather than LangChain's
  BaseMessage. Every caller in this library holds the runtime-neutral Message
  from agent.messages(), so ChatComponent.humanContent carried a cast and the
  docs told consumers to pass something other than the message they actually
  have. The function reads nothing but .content, so the structural parameter is
  the honest signature; the cast is gone.

- --tplane-chat-launcher-offset-x / -y (both 1rem) position the <chat-popup>
  launcher, which was pinned with hard-coded corner offsets. The popup window
  reads the horizontal one too, so it stays aligned when the launcher moves
  clear of a bottom bar or a consent banner.

The new message-utils spec fails to type-check against the old BaseMessage
signature (four errors under tsconfig.spec.json) and passes after. The new
chat-popup styles spec fails four of its five cases before the change, and its
window-alignment guard was mutation-checked by reverting that one declaration.

Regenerates the chat and render api-docs.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Author: blove

apps/website/content/docs/chat/api/api-docs.json

@@ -1925,7 +1925,7 @@
       },
       {
         "name": "messageContent",
-        "type": "(message: BaseMessage<>) => string",
+        "type": "(message: object) => string",
         "description": "",
         "optional": false
       },
@@ -2072,12 +2072,12 @@
       },
       {
         "name": "humanContent",
-        "signature": "humanContent(message: unknown): string",
+        "signature": "humanContent(message: object): string",
         "description": "Renderable content for a human-role message bubble. Most human\nmessages are typed prompts and pass through `messageContent`\nunchanged. A2UI action messages (e.g. form submits, button clicks\non a rendered surface) flow through the same submit channel and\nland in the message stream as a HumanMessage whose content is a\nJSON-serialized `A2uiActionMessage`. Showing the raw JSON as if\nthe user typed it leaks the protocol; per the A2UI spec\nthose events resemble tool calls more than user utterances.\n\n`a2uiActionLabel` returns a short human-readable label for\nrecognized action shapes (\"Search flights\", \"Selected flight UA123\",\netc.) — or null for any non-action content, in which case we fall\nback to the original text.",
         "params": [
           {
             "name": "message",
-            "type": "unknown",
+            "type": "object",
             "description": "",
             "optional": false
           }
@@ -4002,7 +4002,7 @@
       {
         "name": "resolvedRegistry",
         "type": "Signal<Readonly<Record<string, Type<unknown> | RenderViewEntry>>>",
-        "description": "",
+        "description": "Most specific wins: the `[viewRegistry]` input, then a registry provided by\nan ancestor injector, then the built-in markdown views.",
         "optional": false
       },
       {
@@ -6467,32 +6467,6 @@
     ],
     "examples": []
   },
-  {
-    "name": "ChatConfig",
-    "kind": "interface",
-    "description": "Application-wide options for provideChat. Every field is optional;\nthe values are exposed to all chat components in the tree via the\n`CHAT_CONFIG` injection token, so you set them once at bootstrap instead of\nthreading props through every component.",
-    "properties": [
-      {
-        "name": "assistantName",
-        "type": "string",
-        "description": "Shared assistant display name for consumers that read CHAT_CONFIG (default: \"Assistant\").",
-        "optional": true
-      },
-      {
-        "name": "avatarLabel",
-        "type": "string",
-        "description": "Shared AI avatar label for consumers that read CHAT_CONFIG (default: \"A\").",
-        "optional": true
-      },
-      {
-        "name": "renderRegistry",
-        "type": "AngularRegistry",
-        "description": "Shared render registry for consumers that read CHAT_CONFIG.",
-        "optional": true
-      }
-    ],
-    "examples": []
-  },
   {
     "name": "ChatLifecycle",
     "kind": "interface",
@@ -8468,7 +8442,7 @@
     "name": "ContentType",
     "kind": "type",
     "description": "",
-    "signature": "\"pending\" | \"markdown\" | \"json-render\" | \"a2ui\" | \"mixed\"",
+    "signature": "\"pending\" | \"markdown\" | \"json-render\" | \"a2ui\"",
     "examples": []
   },
   {
@@ -8590,13 +8564,6 @@
     "signature": "ViewRegistry",
     "examples": []
   },
-  {
-    "name": "CHAT_CONFIG",
-    "kind": "const",
-    "description": "",
-    "signature": "InjectionToken<ChatConfig>",
-    "examples": []
-  },
   {
     "name": "CHAT_LIFECYCLE",
     "kind": "const",
@@ -8614,7 +8581,7 @@
   {
     "name": "MARKDOWN_VIEW_REGISTRY",
     "kind": "const",
-    "description": "DI token for the markdown view registry consumed by <chat-streaming-md>\nand <md-children>. Maps MarkdownNode.type strings (e.g. \"paragraph\",\n\"heading\") to Angular components that render that node type.\n\n`<chat-streaming-md>` provides the runtime registry on its component-level\ninjector — either the consumer-supplied [viewRegistry] input, or\n`cacheplaneMarkdownViews` (the default) — so descendant <md-children>\ncomponents resolve the right components for each node.",
+    "description": "DI token for the markdown view registry consumed by <chat-streaming-md>\nand <md-children>. Maps MarkdownNode.type strings (e.g. \"paragraph\",\n\"heading\") to Angular components that render that node type.\n\n`<chat-streaming-md>` provides the resolved registry on its component-level\ninjector so descendant <md-children> components resolve the right component\nfor each node. It resolves most-specific-first: the `[viewRegistry]` input,\nthen a registry provided by an ancestor injector (application root or route),\nthen `cacheplaneMarkdownViews` (the default). Providing this token at the\napplication root is therefore a supported app-wide override.",
     "signature": "InjectionToken<Readonly<Record<string, Type<unknown> | RenderViewEntry>>>",
     "examples": []
   },
@@ -9418,13 +9385,13 @@
   {
     "name": "messageContent",
     "kind": "function",
-    "description": "Extracts a human-readable string from a message's content.\n\n`BaseMessage.content` is `string | MessageContentComplex[]`. Reasoning-\ncapable models (OpenAI gpt-5/o-series, Anthropic) emit complex arrays of\ntyped blocks: `{type:'text',text}`, `{type:'reasoning',...}`, tool-use\nblocks, etc. We render only the visible text portions and skip anything\nelse. Stringifying the whole array would dump raw JSON like\n`[{\"type\":\"text\",...}]` into the chat bubble.",
-    "signature": "messageContent(message: BaseMessage<>): string",
+    "description": "Extracts a human-readable string from a message's content.\n\nMessage content is either a plain string or an array of typed blocks.\nReasoning-capable models (OpenAI gpt-5/o-series, Anthropic) emit complex\narrays: `{type:'text',text}`, `{type:'reasoning',...}`, tool-use blocks, etc.\nOnly the visible text portions are rendered and anything else is skipped.\nStringifying the whole array would dump raw JSON like `[{\"type\":\"text\",...}]`\ninto the chat bubble.\n\nThe parameter is structural on purpose. This function reads nothing but\n`.content`, and callers hold either the runtime-neutral `Message` from\n`agent.messages()` or a LangChain `BaseMessage` depending on where the\nmessage came from. Both satisfy `{ content: unknown }`, so neither has to\ncast.",
+    "signature": "messageContent(message: object): string",
     "params": [
       {
         "name": "message",
-        "type": "BaseMessage<>",
-        "description": "",
+        "type": "object",
+        "description": "Any object carrying a `content` field.",
         "optional": false
       }
     ],
@@ -9531,27 +9498,6 @@
     },
     "examples": []
   },
-  {
-    "name": "provideChat",
-    "kind": "function",
-    "description": "Bootstrap `@threadplane/chat` in an Angular application or standalone\ncomponent tree.\n\nCall this once inside `bootstrapApplication` (or the `providers` array of a\nroot `ApplicationConfig`). It registers the shared ChatConfig token\nso every chat component in the tree can read the render registry, avatar\nlabel, and assistant display name without explicit prop threading.",
-    "signature": "provideChat(config: ChatConfig): EnvironmentProviders",
-    "params": [
-      {
-        "name": "config",
-        "type": "ChatConfig",
-        "description": "Options bag that controls the chat feature set:\n  - `renderRegistry` — shared AngularRegistry wiring tool-view\n    components to their names; pass the value returned by\n    `defineAngularRegistry` from `\\@threadplane/render`.\n  - `avatarLabel` — short label shown in the AI avatar bubble (default `\"A\"`).\n  - `assistantName` — display name shown above assistant messages\n    (default `\"Assistant\"`).",
-        "optional": false
-      }
-    ],
-    "returns": {
-      "type": "EnvironmentProviders",
-      "description": ""
-    },
-    "examples": [
-      "```ts\n// main.ts\nimport { bootstrapApplication } from '@angular/platform-browser';\nimport { provideChat } from '@threadplane/chat';\nimport { defineAngularRegistry, provideRender } from '@threadplane/render';\nimport { DayCardComponent } from './day-card.component';\n\nconst registry = defineAngularRegistry({ day_card: DayCardComponent });\n\nbootstrapApplication(AppComponent, {\n  providers: [\n    provideChat({ renderRegistry: registry, avatarLabel: 'AI' }),\n    provideRender({ registry }),\n  ],\n});\n```"
-    ]
-  },
   {
     "name": "renderMarkdown",
     "kind": "function",
@@ -9771,7 +9717,7 @@
   {
     "name": "tools",
     "kind": "function",
-    "description": "Collect named client tools into a frozen, name-keyed registry.\n\nThe overload is generic over the entire map (`const M`) so that each tool's\nprecise type (FunctionToolDef`<S,R>`, ViewToolDef`<S,C>`, or\nAskToolDef`<S,C>`) and every literal key are preserved in the\nClientToolRegistry passed to `provideChat`. This lets downstream\nconsumers look up individual tools without losing generic information.",
+    "description": "Collect named client tools into a frozen, name-keyed registry.\n\nThe overload is generic over the entire map (`const M`) so that each tool's\nprecise type (FunctionToolDef`<S,R>`, ViewToolDef`<S,C>`, or\nAskToolDef`<S,C>`) and every literal key are preserved in the\nClientToolRegistry passed to the `clientTools` input. This lets downstream\nconsumers look up individual tools without losing generic information.",
     "signature": "tools(map: M): Readonly<M>",
     "params": [
       {

apps/website/content/docs/chat/api/content-classifier.mdx

@@ -60,7 +60,7 @@ interface ContentClassifier {
 ## ContentType
 
 ```typescript
-type ContentType = 'pending' | 'markdown' | 'json-render' | 'a2ui' | 'mixed';
+type ContentType = 'pending' | 'markdown' | 'json-render' | 'a2ui';
 ```
 
 | Value | Meaning |
@@ -69,7 +69,8 @@ type ContentType = 'pending' | 'markdown' | 'json-render' | 'a2ui' | 'mixed';
 | `markdown` | Plain text / markdown prose |
 | `json-render` | JSON spec detected (first non-whitespace is `{`) |
 | `a2ui` | A2UI payload detected via `---a2ui_JSON---` prefix, parsed as JSONL messages |
-| `mixed` | Reserved. The current implementation never emits this value. |
+
+The union has no member for interleaved content: prose with inline JSON-render specs classifies as `markdown` and the markdown path renders the embedded specs in place.
 
 ### a2uiSurfaces
 

apps/website/content/docs/chat/components/chat-message-list.mdx

@@ -121,15 +121,13 @@ const type = getMessageType(message); // 'human' | 'ai' | 'tool' | 'system' | 'f
 Runtime-neutral messages have a `content` property that is either a `string` or a `ContentBlock[]`. The library exports a `messageContent()` utility that flattens either shape to a string:
 
 ```typescript
-import type { BaseMessage } from '@langchain/core/messages';
-
 // If content is a string, returns it directly.
 // If content is a block array, concatenates the visible text blocks
 // (`text` and `output_text`) and skips reasoning, tool-use, and image blocks.
-function messageContent(message: BaseMessage): string
+function messageContent(message: { content: unknown }): string
 ```
 
-Note the parameter type: `messageContent()` consumes the LangChain `BaseMessage` shape, not the runtime-neutral `Message` handed to a `chatMessageTemplate`. Pass it a message from a LangChain-shaped source, not one straight out of `agent.messages()`.
+The parameter is structural: the function reads nothing but `content`. Pass it the `Message` handed to a `chatMessageTemplate` straight out of `agent.messages()`, or a LangChain `BaseMessage` from a LangChain-shaped source. Both type-check, and neither needs a cast.
 
 For custom templates, access `message.content` directly and narrow the type in the component class. Angular template expressions have no `typeof` operator, so the check has to live in a method:
 

apps/website/content/docs/chat/components/chat-popup.mdx

@@ -126,13 +126,22 @@ Project content into the window header with the `[chatHeader]` slot:
 
 ## Styling
 
-The popup uses the standard `--tplane-chat-*` token system. The launcher position is not tokenized: the host is `position: fixed; bottom: 1rem; right: 1rem`, so move it by overriding those properties on the host element from a global stylesheet:
+The popup uses the standard `--tplane-chat-*` token system. Two tokens control where the launcher sits, both defaulting to `1rem`:
+
+| Token | Default | Controls |
+|-------|---------|----------|
+| `--tplane-chat-launcher-offset-x` | `1rem` | Distance from the right edge of the viewport |
+| `--tplane-chat-launcher-offset-y` | `1rem` | Distance from the bottom edge of the viewport |
+
+Set them anywhere the popup inherits from, such as `:root`, to clear a bottom bar or a consent banner:
 
 ```css
-chat-popup {
-  bottom: 1.5rem;
-  right: 1.5rem;
+:root {
+  --tplane-chat-launcher-offset-x: 1.5rem;
+  --tplane-chat-launcher-offset-y: 5rem;
 }
 ```
 
+The popup window reads the same horizontal offset, so it stays aligned with the launcher when you move it. Below 640px the window goes full screen and ignores both.
+
 See [Theming](/docs/chat/guides/theming) for the full token reference.

apps/website/content/docs/chat/concepts/message-model.mdx

@@ -164,7 +164,7 @@ type ContentBlock =
 
 Plain text is the common case, and it is the only case in the demo: the LangGraph adapter flattens LangChain content arrays down to their visible text before the message reaches a component, keeping the raw payload on `extra`. Adapters that can preserve more shape emit blocks instead.
 
-Custom templates should therefore check the type before assuming they can interpolate `content` directly. `messageContent()` does this for you, taking a LangChain `BaseMessage` rather than the `Message` shape, and hand-written code can do the same:
+Custom templates should therefore check the type before assuming they can interpolate `content` directly. `messageContent()` does this for you and accepts the `Message` shape directly, and hand-written code can do the same:
 
 ```ts
 function textOf(message: Message): string {

apps/website/content/docs/chat/guides/streaming.mdx

@@ -20,7 +20,7 @@ Each AI message is processed by a `ContentClassifier` that examines the content
 | Prose with inline JSON-render specs | `markdown` | Markdown path, with embedded specs rendered in place |
 | Any other text | `markdown` | Rendered as markdown prose |
 
-Prose that interleaves inline JSON-render specs still classifies as `'markdown'` — the markdown path renders the embedded specs in place. The `ContentType` union also includes `'mixed'`, but `createContentClassifier` does not currently emit it, so treat it as reserved: do not branch on `classifier.type() === 'mixed'` expecting inline-spec content to land there. (`'a2ui'` is covered under [A2UI Content Detection](#a2ui-content-detection) below.)
+Prose that interleaves inline JSON-render specs still classifies as `'markdown'` — the markdown path renders the embedded specs in place. There is no separate type for interleaved content. (`'a2ui'` is covered under [A2UI Content Detection](#a2ui-content-detection) below.)
 
 <Callout type="info" title="Per-message classification">
 Each message gets its own classifier instance. Classification happens once per message — the type is determined by the first meaningful character and never changes.
@@ -68,7 +68,7 @@ classifier.dispose();
 
 | Signal | Type | Description |
 |--------|------|-------------|
-| `type` | `Signal<ContentType>` | `'pending'`, `'markdown'`, `'json-render'`, `'a2ui'`, or `'mixed'` |
+| `type` | `Signal<ContentType>` | `'pending'`, `'markdown'`, `'json-render'`, or `'a2ui'` |
 | `markdown` | `Signal<string>` | Accumulated markdown prose (empty for pure JSON) |
 | `spec` | `Signal<Spec \| null>` | Materialized JSON-render spec with structural sharing |
 | `elementStates` | `Signal<Map<string, ElementAccumulationState>>` | Per-element tracking of which properties have been received |
@@ -77,7 +77,7 @@ classifier.dispose();
 ### ContentType
 
 ```typescript
-type ContentType = 'pending' | 'markdown' | 'json-render' | 'a2ui' | 'mixed';
+type ContentType = 'pending' | 'markdown' | 'json-render' | 'a2ui';
 ```
 
 ## Using ParseTreeStore Directly

apps/website/content/docs/chat/guides/theming.mdx

@@ -94,6 +94,8 @@ Three shadow tokens, `--tplane-chat-shadow-sm`, `--tplane-chat-shadow-md`, and `
 | `--tplane-chat-radius-card` | `8px` | Tool call cards, citations, generative UI surfaces |
 | `--tplane-chat-radius-button` | `8px` | Buttons in panels and lists |
 | `--tplane-chat-radius-launcher` | `9999px` | Circular launcher button |
+| `--tplane-chat-launcher-offset-x` | `1rem` | Distance from the viewport right edge to the `<chat-popup>` launcher and window |
+| `--tplane-chat-launcher-offset-y` | `1rem` | Distance from the viewport bottom edge to the `<chat-popup>` launcher |
 | `--tplane-chat-max-width` | `48rem` | Message column and input width |
 | `--tplane-chat-edge-pad` | `16px` | Horizontal padding at the edges of the conversation |
 | `--tplane-chat-space-1` … `--tplane-chat-space-6`, `--tplane-chat-space-8` | `4px` … `24px`, `32px` | Internal spacing scale |