anyclaude-sdk 0.7.1 → 0.7.3

package/README.md

@@ -358,9 +358,43 @@ npm install anyclaude-react
 
 `useAgent()` plus restylable components — chat (`AgentChat`, `ChatPanel`, `Transcript`, `MarkdownMessage`, `Composer`, `Working`, `ToolCall`) and an IDE set (`Terminal`, `FileExplorer`, `CodeEditor`, `AskUser`). `createAgentClient` / `createEndpointClient` auto-stitch `paused` continuations and run `clientTools` in the browser.
 
+## Run Claude Code against any model — `anyclaude-sdk/anthropic-endpoint`
+
+Stand up an Anthropic Messages API-compatible endpoint backed by any OpenAI-compatible model, so **Claude Code itself** (or any Anthropic-Messages client) runs against DeepSeek / Qwen / GLM / Kimi / local Ollama. Unlike a naive proxy, inline tool-call **dialects are recovered into proper `tool_use` blocks**, so tool use actually works on cheap models.
+
+```ts
+import { createOpenAIClient } from 'anyclaude-sdk/llm'
+import { anthropicToChat, anthropicSSE } from 'anyclaude-sdk/anthropic-endpoint'
+
+const llm = createOpenAIClient({ baseUrl: 'https://api.deepseek.com/v1', model: 'deepseek-chat', apiKey })
+// POST /v1/messages:
+for await (const evt of anthropicSSE(llm, anthropicToChat(body), { model: 'deepseek-chat' })) res.write(evt)
+// then: ANTHROPIC_BASE_URL=http://localhost:8787 claude
+```
+
+Runnable: [`examples/claude-code-router`](examples/claude-code-router).
+
+## Reliable tool use on cheap / open models
+
+Frontier models emit clean native function-calls; cheaper ones often don't. Three layers (in `anyclaude-sdk/llm`) close the gap: **tool-call dialects** (`parseToolCalls` — xml-function / hermes / json-fence), **auto-detected model profiles** (`profileForModel` — qwen/deepseek/moonshot/zhipu/mistral/llama), and **self-healing argument repair** (`query({ repairToolCalls })`, on by default — validates args and feeds the model a corrective tool_result instead of running with garbage). Prove it on your endpoints with [`scripts/compat-matrix.mjs`](scripts/compat-matrix.mjs) → [COMPATIBILITY.md](COMPATIBILITY.md).
+
+## Scaffold an in-browser AI IDE
+
+```bash
+npm create anyclaude-app@latest my-app   # template: bolt — WebContainer + chat + live preview, no backend
+```
+
+The `bolt` template wires `useWebContainerPreview({ wc })` (boot a dev server → live preview URL) + a browser-side `query()` + the IDE components. See [`anyclaude-react`](#react-ui-kit--anyclaude-react).
+
+## Other niceties
+
+- **Live compaction marker** — `autoCompact` emits a `compact_boundary` with `status: 'start'` *before* summarizing (for a live "compacting…" shimmer) and `status: 'end'` after with `post_tokens`.
+- **Cancel a queued message** — `MessageQueue.push()` returns a stable id; `remove(id)` cancels a single pending message (per-pill ✕ in a UI).
+- **BYO LLM client** — reuse the SDK's wire codec: `toOpenAIMessages`, `consumeSSE`, and the LLM types from `anyclaude-sdk/llm` (no bare-root import in browser bundles).
+
 ## Examples & live demo
 
-Runnable Vite projects in [`examples/`](examples/): **`browser-ide`** (WebContainer IDE — real shell + Node in the tab), `browser-chat`, `vercel-kv-survivor`, `vercel-supabase-survivor`, `vercel-indexeddb-survivor`, **`vercel-clienttools`** (server brain / browser hands). Try the **[live demo](https://anyclaude-docs.puter.site/demo/)**.
+Runnable Vite projects in [`examples/`](examples/): **`browser-ide`** (WebContainer IDE — real shell + Node in the tab), `browser-chat`, `claude-code-router`, `vercel-kv-survivor`, `vercel-supabase-survivor`, `vercel-indexeddb-survivor`, **`vercel-clienttools`** (server brain / browser hands). Try the **[live demo](https://anyclaude-docs.puter.site/demo/)**.
 
 ## API
 
@@ -376,6 +410,10 @@ Runnable Vite projects in [`examples/`](examples/): **`browser-ide`** (WebContai
 - `WebContainerWorkspace`, `MemoryFileSystem`, `NoopCommandExecutor`, `LocalSandbox`, `composeWorkspace`
 - `defineTool` (custom tools), `projectMessages` (server-side stream redaction)
 - `ALL_CLAUDE_CODE_TOOLS`, individual tools, `toolDefs`, `toolByName`
+- browser-clean subpaths: `anyclaude-sdk/{query,loop,llm,fs,workspace,tools,session,memory,compact,permissions,skills,queue,prompt,anthropic-endpoint,telemetry}`
+- `anyclaude-sdk/llm`: `parseToolCalls` + dialects, `profileForModel` (model profiles), `validateToolArguments` (repair), `toOpenAIMessages` / `consumeSSE` (BYO-client codec)
+- `anyclaude-sdk/anthropic-endpoint`: `anthropicToChat`, `anthropicSSE`, `streamResultToAnthropicMessage` (Claude-Code router)
+- `runToolLoop` (`/loop`), `compactWithWindow` (`/compact`), `track`/`telemetryEnabled` (`/telemetry`)
 - All `SDK*` message types, `ContentBlockParam`, `LLMClient`, `ToolDef`, `SessionStoreLike`, etc.
 
 ## Differences from the official SDK
@@ -392,7 +430,7 @@ Runnable Vite projects in [`examples/`](examples/): **`browser-ide`** (WebContai
 
 ## Telemetry
 
-The SDK emits **anonymous, opt-out** usage telemetry (SDK version, runtime, a coarse model-family bucket, and which features are used) — never code, prompts, repo identity, paths, or keys, and a no-op unless you configure a collector. Disable with `ANYCLAUDE_TELEMETRY=0`, `DO_NOT_TRACK=1`, or `query({ disableTelemetry: true })`. Full disclosure: [TELEMETRY.md](TELEMETRY.md).
+The SDK emits **anonymous, opt-out** usage telemetry (SDK version, runtime, a coarse model-family bucket, and which features are used) — never code, prompts, repo identity, paths, or keys. It sends to an aggregate-only collector (a Puter Worker; source in [`examples/telemetry-collector`](examples/telemetry-collector)). Disable with `ANYCLAUDE_TELEMETRY=0`, `DO_NOT_TRACK=1`, or `query({ disableTelemetry: true })`; repoint with `ANYCLAUDE_TELEMETRY_URL` (or set it to `''` to send nowhere). Full disclosure: [TELEMETRY.md](TELEMETRY.md).
 
 ## License
 

package/dist/query.js

@@ -3,7 +3,7 @@
 // Returns an AsyncGenerator<SDKMessage>. Accepts either a single string prompt
 // or an async iterable of SDKUserMessage (for multi-turn / interactive use).
 import { runAgent } from './agent.js';
-import { track, telemetryEnabled } from './telemetry.js';
+import { track, telemetryEnabled, tokenBucket } from './telemetry.js';
 import { profileForModel } from './llm/profiles.js';
 export function query(options) {
     const prompt = typeof options.prompt === 'string'
@@ -59,17 +59,18 @@ export function query(options) {
         settings: options.settings,
         skills: options.skills,
     });
-    gen.interrupt = () => abortController.abort();
-    // Anonymous, aggregate adoption signal — one event per public run. Fire-and-forget,
-    // never blocks the generator, no-ops unless enabled + a collector is configured.
-    // Only booleans + a coarse model-family bucket leave the process (see telemetry.ts).
+    // Anonymous, aggregate adoption signal. Fire-and-forget, never blocks, no-ops
+    // unless enabled + a collector is configured. Only booleans + coarse buckets
+    // (model family, token-volume bucket) ever leave the process — see telemetry.ts.
     const telemetry = {
         disabled: options.disableTelemetry,
         ...options.telemetry,
     };
-    if (telemetryEnabled(telemetry)) {
+    const modelFamily = profileForModel(options.model).name;
+    const enabled = telemetryEnabled(telemetry);
+    if (enabled) {
         track('run', {
-            model_family: profileForModel(options.model).name,
+            model_family: modelFamily,
             client_workspace_tools: !!options.clientWorkspaceTools,
             client_tools: !!options.clientTools?.length,
             survivor: options.maxDurationMs != null,
@@ -83,7 +84,29 @@ export function query(options) {
             resumed: !!options.continueRun || !!options.resume,
         }, telemetry);
     }
-    return gen;
+    if (!enabled) {
+        gen.interrupt = () => abortController.abort();
+        return gen;
+    }
+    // Wrap to emit one `run_end` with a coarse token-volume bucket when the run
+    // finishes (tokens aren't known until the `result` message). Pass-through only.
+    const wrapped = (async function* () {
+        let totalTokens = 0;
+        try {
+            for await (const m of gen) {
+                if (m.type === 'result' && m.usage) {
+                    const u = m.usage;
+                    totalTokens = (u.input_tokens || 0) + (u.output_tokens || 0);
+                }
+                yield m;
+            }
+        }
+        finally {
+            track('run_end', { model_family: modelFamily, tokens_bucket: tokenBucket(totalTokens) }, telemetry);
+        }
+    })();
+    wrapped.interrupt = () => abortController.abort();
+    return wrapped;
 }
 /** Wrap a single text prompt into the async-iterable form runAgent expects. */
 export async function* singlePrompt(text) {

package/dist/telemetry.d.ts

@@ -6,6 +6,8 @@ export interface TelemetryOptions {
     /** Collector URL. Defaults to `ANYCLAUDE_TELEMETRY_URL` then the built-in default. */
     url?: string;
 }
+/** Coarse token-volume bucket — never an exact count, so a single run isn't fingerprintable. */
+export declare function tokenBucket(total: number): string;
 /** Resolve whether telemetry may run, honoring every documented opt-out. */
 export declare function telemetryEnabled(opts?: TelemetryOptions): boolean;
 /** Coarse runtime bucket — never anything machine-identifying. */

package/dist/telemetry.js

@@ -24,7 +24,21 @@ const DEFAULT_TELEMETRY_URL = 'https://anyclaude-telemetry.puter.work';
 // Only these prop keys are ever transmitted, and only with safe value types.
 // Booleans pass through; these specific string keys pass through as-is (they are
 // coarse buckets we set ourselves — never free-form / user data).
-const ALLOWED_STRING_KEYS = new Set(['model_family', 'event_detail']);
+const ALLOWED_STRING_KEYS = new Set(['model_family', 'event_detail', 'tokens_bucket']);
+/** Coarse token-volume bucket — never an exact count, so a single run isn't fingerprintable. */
+export function tokenBucket(total) {
+    if (!Number.isFinite(total) || total <= 0)
+        return '0';
+    if (total < 1_000)
+        return '<1k';
+    if (total < 10_000)
+        return '1k-10k';
+    if (total < 100_000)
+        return '10k-100k';
+    if (total < 1_000_000)
+        return '100k-1m';
+    return '1m+';
+}
 function readEnv(name) {
     const p = globalThis.process;
     return p?.env?.[name];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "anyclaude-sdk",
-  "version": "0.7.1",
+  "version": "0.7.3",
   "description": "Standalone, browser-compatible SDK providing Claude Code agent capabilities (tools, tool loop, multi-turn, MCP, sub-agents, sessions) against any OpenAI/Anthropic-compatible LLM endpoint. Runs in the browser (WebContainer), Node, and Bun — no backend required.",
   "type": "module",
   "main": "./dist/index.js",