@d4y/agent-runtime-nuxt 0.1.3 → 0.1.4

package/README.md

@@ -108,7 +108,7 @@ The composable returns AI-SDK / Nuxt UI compatible `UIMessage` shapes, so
 | `/conversations/:id`                                 | DELETE | Delete a persisted conversation.                          |
 | `/conversations/:id/history`                         | GET    | Return the persisted conversation history.                |
 | `/conversations/:id/stream`                          | GET    | Pipes the upstream SSE stream verbatim.                  |
-| `/conversations/:id/messages`                        | POST   | Append a user turn. Body: `{ content, context? }`.       |
+| `/conversations/:id/messages`                        | POST   | Append a user turn. Body: `{ content, context?, requestOptions? }`. |
 | `/conversations/:id/abort`                           | POST   | Cancel the in-flight agent run.                          |
 | `/conversations/:id/env`                             | POST   | Patch workspace env. Body: `{ env, merge? }`.            |
 | `/conversations/:id/files`                           | GET    | List workspace files.                                    |
@@ -160,9 +160,32 @@ workspace via `POST /conversations/:id/env`, which retriggers any matching
 #### `send(text, options?)`
 
 Lazily calls `start()` if no conversation exists. The displayed user bubble
-always shows `text`; pass `options.rewriteContent` to mutate the *forwarded*
-prompt without changing what the user sees. Useful for model-specific soft
-switches:
+always shows `text`.
+
+Use `requestOptions` for provider-side controls such as Qwen/vLLM
+`chat_template_kwargs`:
+
+```ts
+await chat.start({
+  requestOptions: {
+    temperature: 0.7,
+    topP: 0.8,
+    presencePenalty: 1.5
+  }
+})
+
+await chat.send(input.value, {
+  requestOptions: {
+    extraBody: {
+      top_k: 20,
+      chat_template_kwargs: { enable_thinking: false }
+    }
+  }
+})
+```
+
+`rewriteContent` is still available as a fallback when a model expects prompt
+text toggles rather than provider payload fields:
 
 ```ts
 await chat.send(input.value, {

package/dist/module.json

@@ -4,7 +4,7 @@
   "compatibility": {
     "nuxt": ">=3.13.0"
   },
-  "version": "0.1.3",
+  "version": "0.1.4",
   "builder": {
     "@nuxt/module-builder": "1.0.2",
     "unbuild": "unknown"

package/dist/runtime/composables/useAgentRuntime.d.ts

@@ -38,6 +38,22 @@ export interface UiAction {
     payload: unknown;
     at: number;
 }
+export interface ContextUsage {
+    tokens: number | null;
+    contextWindow: number;
+    percent: number | null;
+}
+export interface CompactionLogEntry {
+    phase: 'start' | 'end';
+    at: number;
+    reason?: string;
+    tokensBefore?: number;
+    tokensAfter?: number;
+    compacted?: boolean;
+    aborted?: boolean;
+    willRetry?: boolean;
+    error?: string;
+}
 /** Workspace file entry, mirrors the backend response. */
 export interface FileEntry {
     name: string;
@@ -47,6 +63,19 @@ export interface FileEntry {
     mimeType: string;
     kind: 'image' | 'html' | 'pdf' | 'download';
 }
+/**
+ * Generic provider request options forwarded to the runtime.
+ * `extraBody` is merged verbatim into the provider payload, which makes this
+ * suitable for model-specific flags such as Qwen's `chat_template_kwargs`.
+ */
+export interface RequestOptions {
+    temperature?: number;
+    maxTokens?: number;
+    topP?: number;
+    presencePenalty?: number;
+    frequencyPenalty?: number;
+    extraBody?: Record<string, unknown>;
+}
 /** Per-turn options for `send`. */
 export interface SendOptions {
     /**
@@ -62,11 +91,15 @@ export interface SendOptions {
     rewriteContent?: (text: string) => string;
     /** Optional language override for this turn / conversation, e.g. `en` or `de-CH`. */
     language?: string;
+    /** Optional provider request overrides for this turn only. */
+    requestOptions?: RequestOptions;
 }
 /** Options for `start`. */
 export interface StartOptions {
     /** Optional language override for the newly created conversation. */
     language?: string;
+    /** Optional default provider request options for the conversation. */
+    requestOptions?: RequestOptions;
 }
 /** Return shape of {@link useAgentRuntime}. All members are reactive. */
 export interface UseAgentRuntime {
@@ -80,6 +113,12 @@ export interface UseAgentRuntime {
     error: Ref<Error | null>;
     /** Most recent UI-action events, newest first, capped at 50. */
     uiActions: Ref<UiAction[]>;
+    /** Latest context usage sample from the active PI session. */
+    contextUsage: Ref<ContextUsage | null>;
+    /** True iff a compaction run is currently in progress. */
+    compactionInProgress: Ref<boolean>;
+    /** Most recent compaction events, newest first. */
+    compactionLog: Ref<CompactionLogEntry[]>;
     /** Active app manifest. `null` until the first `onMounted` fetch resolves. */
     app: Ref<AppInfo | null>;
     /** Current auth values (persisted in localStorage, scoped per appId). */

package/dist/runtime/composables/useAgentRuntime.js

@@ -43,6 +43,9 @@ export const useAgentRuntime = () => {
   const status = ref("idle");
   const error = ref(null);
   const uiActions = ref([]);
+  const contextUsage = ref(null);
+  const compactionInProgress = ref(false);
+  const compactionLog = ref([]);
   const app = ref(null);
   const auth = ref(emptyAuth());
   const authReady = computed(() => computeAuthReady(auth.value, app.value));
@@ -121,6 +124,12 @@ export const useAgentRuntime = () => {
       return { ...msg, parts };
     });
   };
+  const normalizePercent = (value) => Math.max(0, Math.min(100, value > 1 ? value : value * 100));
+  const toNumber = (value) => typeof value === "number" && Number.isFinite(value) ? value : void 0;
+  const toNullableNumber = (value) => value === null ? null : toNumber(value);
+  const pushCompactionLog = (entry) => {
+    compactionLog.value = [entry, ...compactionLog.value].slice(0, 20);
+  };
   const onSseEvent = (eventName, data) => {
     switch (eventName) {
       case "agent_start":
@@ -148,23 +157,66 @@ export const useAgentRuntime = () => {
           type: `tool-${d.toolName}`,
           toolCallId: d.toolCallId,
           state: d.isError ? "output-error" : "output-available",
-          output: d.isError ? void 0 : "ok",
-          errorText: d.isError ? "Tool returned an error" : void 0
+          output: d.isError ? void 0 : d.output,
+          errorText: d.isError ? d.errorText ?? "Tool returned an error" : void 0
         }));
         break;
       }
+      case "context_usage": {
+        const d = data;
+        const tokens = toNullableNumber(d.tokens);
+        const contextWindow = toNumber(d.contextWindow);
+        if (tokens === void 0 || contextWindow === void 0) break;
+        const fallbackPercent = typeof tokens === "number" && contextWindow > 0 ? tokens / contextWindow * 100 : null;
+        const percent = toNullableNumber(d.percent) ?? fallbackPercent;
+        if (percent === void 0) break;
+        contextUsage.value = {
+          tokens,
+          contextWindow,
+          percent: typeof percent === "number" ? normalizePercent(percent) : null
+        };
+        break;
+      }
+      case "compaction_start": {
+        const d = data;
+        compactionInProgress.value = true;
+        pushCompactionLog({
+          phase: "start",
+          at: Date.now(),
+          reason: typeof d.reason === "string" && d.reason.trim().length > 0 ? d.reason.trim() : void 0
+        });
+        break;
+      }
+      case "compaction_end": {
+        const d = data;
+        const result = d.result && typeof d.result === "object" ? d.result : null;
+        compactionInProgress.value = false;
+        pushCompactionLog({
+          phase: "end",
+          at: Date.now(),
+          reason: typeof d.reason === "string" && d.reason.trim().length > 0 ? d.reason.trim() : void 0,
+          compacted: typeof d.aborted === "boolean" ? !d.aborted && !!result : !!result,
+          aborted: typeof d.aborted === "boolean" ? d.aborted : void 0,
+          willRetry: typeof d.willRetry === "boolean" ? d.willRetry : void 0,
+          tokensBefore: toNumber(result?.tokensBefore),
+          error: typeof d.errorMessage === "string" && d.errorMessage.trim().length > 0 ? d.errorMessage : void 0
+        });
+        break;
+      }
       case "ui_action": {
         const d = data;
         uiActions.value = [{ toolCallId: d.toolCallId, type: d.type, payload: d.payload, at: Date.now() }, ...uiActions.value].slice(0, 50);
         break;
       }
       case "end":
+        compactionInProgress.value = false;
         status.value = "ready";
         currentAssistantId = null;
         void refreshFiles();
         break;
       case "error": {
         const d = data;
+        compactionInProgress.value = false;
         error.value = new Error(d?.message ?? "stream error");
         status.value = "error";
         currentAssistantId = null;
@@ -217,7 +269,11 @@ export const useAgentRuntime = () => {
     error.value = null;
     const res = await $fetch(`${apiPrefix}/conversations`, {
       method: "POST",
-      body: { env: buildEnvFromAuth(auth.value), language: options.language }
+      body: {
+        env: buildEnvFromAuth(auth.value),
+        language: options.language,
+        requestOptions: options.requestOptions
+      }
     });
     conversationId.value = res.conversationId;
     abortController?.abort();
@@ -264,7 +320,12 @@ export const useAgentRuntime = () => {
     const wireContent = options.rewriteContent ? options.rewriteContent(text) : text;
     await $fetch(`${apiPrefix}/conversations/${conversationId.value}/messages`, {
       method: "POST",
-      body: { content: wireContent, context: options.context, language: options.language }
+      body: {
+        content: wireContent,
+        context: options.context,
+        language: options.language,
+        requestOptions: options.requestOptions
+      }
     }).catch((err) => {
       error.value = err instanceof Error ? err : new Error(String(err));
       status.value = "error";
@@ -283,6 +344,9 @@ export const useAgentRuntime = () => {
     currentAssistantId = null;
     replaceMessages([]);
     uiActions.value = [];
+    contextUsage.value = null;
+    compactionInProgress.value = false;
+    compactionLog.value = [];
     files.value = [];
     status.value = "idle";
     error.value = null;
@@ -296,6 +360,9 @@ export const useAgentRuntime = () => {
     status,
     error,
     uiActions,
+    contextUsage,
+    compactionInProgress,
+    compactionLog,
     app,
     auth,
     authReady,

package/dist/runtime/server/api/conversations/[id]/messages.post.js

@@ -9,7 +9,12 @@ export default defineEventHandler(async (event) => {
   const response = await fetch(`${cfg.baseUrl}/v1/conversations/${id}/messages`, {
     method: "POST",
     headers: agentRuntimeHeaders(cfg),
-    body: JSON.stringify({ content: body.content, context: body.context, language: body.language })
+    body: JSON.stringify({
+      content: body.content,
+      context: body.context,
+      requestOptions: body.requestOptions,
+      language: body.language
+    })
   });
   if (!response.ok) {
     throw createError({

package/dist/runtime/server/api/conversations.post.js

@@ -6,7 +6,13 @@ export default defineEventHandler(async (event) => {
   const response = await fetch(`${cfg.baseUrl}/v1/conversations`, {
     method: "POST",
     headers: agentRuntimeHeaders(cfg),
-    body: JSON.stringify({ appId: cfg.appId, env: body.env ?? {}, model: body.model, language: body.language })
+    body: JSON.stringify({
+      appId: cfg.appId,
+      env: body.env ?? {},
+      model: body.model,
+      requestOptions: body.requestOptions,
+      language: body.language
+    })
   });
   if (!response.ok) {
     throw createError({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@d4y/agent-runtime-nuxt",
-  "version": "0.1.3",
+  "version": "0.1.4",
   "description": "Headless Nuxt module that connects a Nuxt app to an agent-runtime server. Ships server-side proxy routes (so your X-Agent-Runtime-App-Key never leaves the server) and a single composable, useAgentRuntime(), that exposes a typed chat client.",
   "license": "MIT",
   "type": "module",