feat(ai-anthropic): support native on-demand compaction

.changeset/anthropic-native-compaction.md

@@ -0,0 +1,7 @@
+---
+'@tanstack/ai': minor
+'@tanstack/ai-compaction': minor
+'@tanstack/ai-anthropic': minor
+---
+
+`withCompaction({ native })` now passes the tools, the model options, and the reasoning of the chat call into `compact`. The Anthropic text adapter implements `compact` as an on-demand Messages request (`compaction: { type: "summarize" }`, beta `compact-2026-09-04`). The signed compaction block stays on the assistant message and goes back on later requests to the same model. A result with no signed block uses the configured strategy.

docs/adapters/anthropic-advanced.md

@@ -256,6 +256,38 @@ export const throughProxy = anthropicText("claude-opus-5-5", {
 
 If that endpoint sends the request and the `anthropic-beta` header on to Anthropic without changes, set `true`. See [Mid-Conversation Changes](../advanced/mid-conversation-changes).
 
+## Native compaction
+
+Anthropic can compact a long conversation on its side. `createAnthropicChat` and `anthropicText` have a `compact` method for that call. Pass the adapter as `native`:
+
+```typescript
+import { chat } from "@tanstack/ai";
+import { createAnthropicChat } from "@tanstack/ai-anthropic";
+import { withCompaction } from "@tanstack/ai-compaction";
+
+const adapter = createAnthropicChat(
+  "claude-opus-5-5",
+  process.env.ANTHROPIC_API_KEY!,
+);
+
+const stream = chat({
+  adapter,
+  messages: [{ role: "user", content: "Hello!" }],
+  threadId: "thread-1",
+  middleware: [withCompaction({ maxTokens: 100_000, native: adapter })],
+});
+```
+
+The call is a Messages request. It sets `compaction` to `{ type: "summarize" }` and sends the beta `compact-2026-09-04`.
+
+- The request includes the system prompts, the tools, and `max_tokens` from the chat. The call does not run a tool.
+- The result is one assistant message. The message text is the summary.
+- `metadata.anthropicCompaction` holds the signed block. The block has `content` and `signature`.
+- The next request to the same model includes that block and the same beta.
+- A result with no signed block uses `strategy`. After a 404, a 405, or a 501, that adapter and model use `strategy` until the process restarts.
+
+See [Let the provider compact](../advanced/compaction#let-the-provider-compact).
+
 ## Provider tools
 
 Anthropic has tools of its own, next to the function tools that you define. Import them from `@tanstack/ai-anthropic/tools` and pass them into `chat({ tools: [...] })`.

docs/adapters/anthropic.md

@@ -356,8 +356,8 @@ Now a declined request gets an answer from a fallback model. When no model can a
 These topics are on [Anthropic: Advanced](./anthropic-advanced):
 
 - Bearer and OAuth tokens, gateways, and your own Anthropic client.
-- Claude on Vertex AI.
-- The `max_tokens` default and prompt caching.
+- Claude on Vertex AI, the `max_tokens` default, and prompt caching.
+- [Native compaction](./anthropic-advanced#native-compaction).
 - Every provider tool, with a full sample.
 - The API reference.
 

docs/advanced/compaction-advanced.md

@@ -91,7 +91,7 @@ These rules apply when you pass `native` (see [Let the provider compact](./compa
 - When `compact` fails, `strategy` runs in its place. `onCompact` and `compaction:ended` get the failure in `error`. Some providers on the OpenAI Responses base have no compaction endpoint.
 - After a 404, 405, or 501 (the provider has no compaction endpoint), that adapter object and model use `strategy` until the process restarts. Create the adapter once, outside the request handler, so that no more compaction calls go out.
 - After any other failure, such as a 400, a 429, or a network error, the next compaction tries `compact` again.
-- The compaction call gets the system prompts of the `chat()` call. It does not get the tools, because the compaction endpoint does not accept them.
+- The compaction call gets the system prompts, the tools, the model options, and the reasoning of the `chat()` call.
 - The compacted messages can hold an encrypted provider item. The adapter sends it back unchanged to the same model. Another model does not get it.
 
 ## Give the retry something to cut

docs/advanced/compaction.md

@@ -238,7 +238,18 @@ export async function POST(request: Request) {
 - An adapter with no `compact` method uses `strategy`.
 - When `compact` fails, `strategy` runs in its place.
 
-The OpenAI Responses adapters have `compact`. See [Native compaction](../adapters/openai-advanced#native-compaction). For what happens on each kind of failure, see [Native compaction failures](./compaction-advanced#native-compaction-failures).
+The Anthropic adapters have `compact` too. Create the adapter once:
+
+```typescript
+import { createAnthropicChat } from "@tanstack/ai-anthropic";
+
+const adapter = createAnthropicChat(
+  "claude-opus-5-5",
+  process.env.ANTHROPIC_API_KEY!,
+);
+```
+
+Pass that adapter to `chat()` and to `native`, as in the sample above. See [OpenAI native compaction](../adapters/openai-advanced#native-compaction) and [Anthropic native compaction](../adapters/anthropic-advanced#native-compaction). For each kind of failure, see [Native compaction failures](./compaction-advanced#native-compaction-failures).
 
 ## When the chat is still too long
 

docs/config.json

@@ -450,13 +450,13 @@
           "label": "Compaction",
           "to": "advanced/compaction",
           "addedAt": "2026-08-24",
-          "updatedAt": "2026-10-22"
+          "updatedAt": "2026-11-09"
         },
         {
           "label": "Compaction: Advanced",
           "to": "advanced/compaction-advanced",
           "addedAt": "2026-10-21",
-          "updatedAt": "2026-11-07"
+          "updatedAt": "2026-11-09"
         }
       ],
       "tab": "guides"
@@ -1724,13 +1724,13 @@
           "label": "Anthropic",
           "to": "adapters/anthropic",
           "addedAt": "2026-04-15",
-          "updatedAt": "2026-10-22"
+          "updatedAt": "2026-11-09"
         },
         {
           "label": "Anthropic: Advanced",
           "to": "adapters/anthropic-advanced",
           "addedAt": "2026-10-21",
-          "updatedAt": "2026-11-02"
+          "updatedAt": "2026-11-09"
         },
         {
           "label": "Google Gemini",

docs/reference/interfaces/TextCompactOptions.md

@@ -3,7 +3,7 @@ id: TextCompactOptions
 title: TextCompactOptions
 ---
 
-Defined in: [packages/ai/src/types.ts:1570](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1570)
+Defined in: [packages/ai/src/types.ts:1584](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1584)
 
 What `TextAdapter.compact` receives.
 
@@ -62,6 +62,54 @@ The system prompts of the chat call.
 optional wrapFetch?: FetchWrapper;
 ```
 
-Defined in: [packages/ai/src/types.ts:1576](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1576)
+Defined in: [packages/ai/src/types.ts:1591](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1591)
 
 Wraps the fetch of the compaction request.
+
+***
+
+### tools?
+
+```ts
+optional tools?: AnyTool[];
+```
+
+Defined in: [packages/ai/src/types.ts:1598](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1598)
+
+The tools of the chat call. The provider reads the definitions. The compaction call does not run a tool.
+
+***
+
+### modelOptions?
+
+```ts
+optional modelOptions?: Record<string, unknown>;
+```
+
+Defined in: [packages/ai/src/types.ts:1600](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1600)
+
+Provider model options of the chat call, including `max_tokens`.
+
+***
+
+### reasoning?
+
+```ts
+optional reasoning?: ReasoningRequest;
+```
+
+Defined in: [packages/ai/src/types.ts:1602](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1602)
+
+How hard the model thinks on the chat call.
+
+***
+
+### addUsage?
+
+```ts
+optional addUsage?: (usage: TokenUsage) => void;
+```
+
+Defined in: [packages/ai/src/types.ts:1604](https://github.com/TanStack/ai/blob/main/packages/ai/src/types.ts#L1604)
+
+Report the token usage of the compaction call.