pi-web-search 1.0.0 → 1.0.2

package/README.md

@@ -5,8 +5,7 @@ A pi extension that provides web search and content analysis capabilities powere
 ## Features
 
 - **🔍 Web Search** - Google search with citations
-- **📄 URL Analysis** - Analyze web pages, PDFs, and images (up to 20 URLs)
-- **🎥 YouTube** - Video summaries and Q&A
+- **📄 URL Analysis** - Analyze web pages, PDFs, images, and YouTube videos (up to 20 URLs)
 
 ## Installation
 

package/dist/index.js

@@ -1,6 +1,5 @@
 import { webSearch, WebSearchSchema } from "./web_search.js";
 import { urlContext, UrlContextSchema } from "./url_context.js";
-import { youtubeVideo, YoutubeVideoSchema } from "./youtube_video.js";
 export default function (pi) {
     pi.registerTool({
         name: "web_search",
@@ -12,15 +11,8 @@ export default function (pi) {
     pi.registerTool({
         name: "url_context",
         label: "URL Context",
-        description: "Analyze web pages and documents. Extract data, compare documents, synthesize content. Supports text/html, PDF, images, JSON, CSV. Up to 20 URLs.",
+        description: "Analyze the content of up to 20 public URLs. Supports web pages, documents, images, and YouTube videos.",
         parameters: UrlContextSchema,
         execute: urlContext
     });
-    pi.registerTool({
-        name: "youtube_video",
-        label: "YouTube Video",
-        description: "Analyze YouTube videos. Summarize, answer questions, find timestamps. Supports video clipping (start/end offsets). Preview feature.",
-        parameters: YoutubeVideoSchema,
-        execute: youtubeVideo
-    });
 }

package/dist/url_context.js

@@ -4,11 +4,12 @@ import { getModel, missingConfigResult, errorResult, formatResult } from "./util
 export const UrlContextSchema = Type.Object({
     query: Type.String({ description: "Question or task to perform on the URLs" }),
     urls: Type.Array(Type.String(), {
-        description: "URLs to analyze (up to 20). Supports text/html, PDF, images, JSON, CSV, etc.",
+        description: "Public URLs to analyze (web pages, documents, images, YouTube videos, etc).",
         minItems: 1,
         maxItems: 20
     }),
 });
+const YOUTUBE_REGEX = /^(?:https?:\/\/)?(?:www\.)?(?:youtube\.com\/(?:watch\?v=|embed\/)|youtu\.be\/)([a-zA-Z0-9_-]{11})/;
 export async function urlContext(id, params, signal, onUpdate, ctx) {
     const model = await getModel(ctx);
     if (!model)
@@ -17,10 +18,56 @@ export async function urlContext(id, params, signal, onUpdate, ctx) {
     onUpdate?.({ content: [{ type: "text", text: `Analyzing ${count} URL${count > 1 ? 's' : ''}...` }], details: {} });
     try {
         const config = getConfig(model);
-        const combinedPrompt = `${params.query}\n\nURLs:\n${params.urls.join("\n")}`;
+        let contents = [];
+        let tools = [{ [config.urlContextTool]: {} }];
+        // Special handling for YouTube videos on Gemini
+        if (model.api === "google-generative-ai") {
+            const youtubeUrls = [];
+            const otherUrls = [];
+            for (const url of params.urls) {
+                if (YOUTUBE_REGEX.test(url)) {
+                    youtubeUrls.push(url);
+                }
+                else {
+                    otherUrls.push(url);
+                }
+            }
+            // If we have YouTube URLs, construct file_data parts
+            if (youtubeUrls.length > 0) {
+                const parts = [];
+                for (const url of youtubeUrls) {
+                    parts.push({
+                        file_data: { file_uri: url, mime_type: "video/mp4" }
+                    });
+                }
+                let prompt = params.query;
+                if (otherUrls.length > 0) {
+                    prompt += `\n\nURLs:\n${otherUrls.join("\n")}`;
+                }
+                else {
+                    // If no other URLs, we might not need the tool, but keep it just in case
+                    // or maybe the tool is required for grounding even with video?
+                    // "google_search_retrieval" tool might confuse if there are no URLs to retrieve.
+                    // But if we remove the tool, we might lose grounding capabilities (like search).
+                    // Let's keep the tool enabled.
+                }
+                parts.push({ text: prompt });
+                contents = [{ role: "user", parts }];
+            }
+            else {
+                // No YouTube URLs, standard behavior
+                const combinedPrompt = `${params.query}\n\nURLs:\n${params.urls.join("\n")}`;
+                contents = [{ role: "user", parts: [{ text: combinedPrompt }] }];
+            }
+        }
+        else {
+            // Not Gemini, standard behavior
+            const combinedPrompt = `${params.query}\n\nURLs:\n${params.urls.join("\n")}`;
+            contents = [{ role: "user", parts: [{ text: combinedPrompt }] }];
+        }
         const result = await callApiStream(ctx, model, {
-            contents: [{ role: "user", parts: [{ text: combinedPrompt }] }],
-            tools: [{ [config.urlContextTool]: {} }]
+            contents,
+            tools
         }, onUpdate);
         const { text, sources } = applyCitations(result.text, result.groundingMetadata);
         // Handle both camelCase and snake_case metadata

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-web-search",
-  "version": "1.0.0",
+  "version": "1.0.2",
   "description": "Web search and content analysis extension for pi, powered by Gemini API",
   "type": "module",
   "main": "dist/index.js",
@@ -24,15 +24,9 @@
     "url": "https://github.com/ttttmr/pi-web-search/issues"
   },
   "homepage": "https://github.com/ttttmr/pi-web-search#readme",
-  "scripts": {
-    "build": "tsc",
-    "prepublishOnly": "npm run build"
-  },
-  "dependencies": {
-    "@mariozechner/pi-coding-agent": "^0.51.5",
-    "@sinclair/typebox": "^0.32.15"
-  },
   "devDependencies": {
+    "@mariozechner/pi-coding-agent": "^0.51.5",
+    "@sinclair/typebox": "^0.32.15",
     "typescript": "^5.0.0",
     "@types/node": "^20.0.0"
   },

package/src/index.ts

@@ -1,7 +1,6 @@
 import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
 import { webSearch, WebSearchSchema } from "./web_search.js";
 import { urlContext, UrlContextSchema } from "./url_context.js";
-import { youtubeVideo, YoutubeVideoSchema } from "./youtube_video.js";
 
 export default function (pi: ExtensionAPI) {
     pi.registerTool({
@@ -15,16 +14,8 @@ export default function (pi: ExtensionAPI) {
     pi.registerTool({
         name: "url_context",
         label: "URL Context",
-        description: "Analyze web pages and documents. Extract data, compare documents, synthesize content. Supports text/html, PDF, images, JSON, CSV. Up to 20 URLs.",
+        description: "Analyze the content of up to 20 public URLs. Supports web pages, documents, images, and YouTube videos.",
         parameters: UrlContextSchema,
         execute: urlContext
     });
-
-    pi.registerTool({
-        name: "youtube_video",
-        label: "YouTube Video",
-        description: "Analyze YouTube videos. Summarize, answer questions, find timestamps. Supports video clipping (start/end offsets). Preview feature.",
-        parameters: YoutubeVideoSchema,
-        execute: youtubeVideo
-    });
 }

package/src/url_context.ts

@@ -6,13 +6,15 @@ import { getModel, missingConfigResult, errorResult, formatResult } from "./util
 export const UrlContextSchema = Type.Object({
     query: Type.String({ description: "Question or task to perform on the URLs" }),
     urls: Type.Array(Type.String(), { 
-        description: "URLs to analyze (up to 20). Supports text/html, PDF, images, JSON, CSV, etc.",
+        description: "Public URLs to analyze (web pages, documents, images, YouTube videos, etc).",
         minItems: 1,
         maxItems: 20
     }),
 });
 export type UrlContextInput = Static<typeof UrlContextSchema>;
 
+const YOUTUBE_REGEX = /^(?:https?:\/\/)?(?:www\.)?(?:youtube\.com\/(?:watch\?v=|embed\/)|youtu\.be\/)([a-zA-Z0-9_-]{11})/;
+
 export async function urlContext(
     id: string, 
     params: UrlContextInput, 
@@ -28,11 +30,60 @@ export async function urlContext(
 
     try {
         const config = getConfig(model);
-        const combinedPrompt = `${params.query}\n\nURLs:\n${params.urls.join("\n")}`;
+        
+        let contents: any[] = [];
+        let tools: any[] | undefined = [{ [config.urlContextTool]: {} }];
+
+        // Special handling for YouTube videos on Gemini
+        if (model.api === "google-generative-ai") {
+            const youtubeUrls: string[] = [];
+            const otherUrls: string[] = [];
+
+            for (const url of params.urls) {
+                if (YOUTUBE_REGEX.test(url)) {
+                    youtubeUrls.push(url);
+                } else {
+                    otherUrls.push(url);
+                }
+            }
+
+            // If we have YouTube URLs, construct file_data parts
+            if (youtubeUrls.length > 0) {
+                const parts: any[] = [];
+                
+                for (const url of youtubeUrls) {
+                    parts.push({
+                        file_data: { file_uri: url, mime_type: "video/mp4" }
+                    });
+                }
+
+                let prompt = params.query;
+                if (otherUrls.length > 0) {
+                    prompt += `\n\nURLs:\n${otherUrls.join("\n")}`;
+                } else {
+                    // If no other URLs, we might not need the tool, but keep it just in case
+                    // or maybe the tool is required for grounding even with video?
+                    // "google_search_retrieval" tool might confuse if there are no URLs to retrieve.
+                    // But if we remove the tool, we might lose grounding capabilities (like search).
+                    // Let's keep the tool enabled.
+                }
+
+                parts.push({ text: prompt });
+                contents = [{ role: "user", parts }];
+            } else {
+                // No YouTube URLs, standard behavior
+                const combinedPrompt = `${params.query}\n\nURLs:\n${params.urls.join("\n")}`;
+                contents = [{ role: "user", parts: [{ text: combinedPrompt }] }];
+            }
+        } else {
+            // Not Gemini, standard behavior
+            const combinedPrompt = `${params.query}\n\nURLs:\n${params.urls.join("\n")}`;
+            contents = [{ role: "user", parts: [{ text: combinedPrompt }] }];
+        }
 
         const result = await callApiStream(ctx, model, {
-            contents: [{ role: "user", parts: [{ text: combinedPrompt }] }],
-            tools: [{ [config.urlContextTool]: {} }]
+            contents,
+            tools
         }, onUpdate);
 
         const { text, sources } = applyCitations(result.text, result.groundingMetadata);

package/dist/advanced_search.d.ts

@@ -1,8 +0,0 @@
-import type { ExtensionContext, AgentToolUpdateCallback } from "@mariozechner/pi-coding-agent";
-import { type Static } from "@sinclair/typebox";
-export declare const AdvancedSearchSchema: import("@sinclair/typebox").TObject<{
-    query: import("@sinclair/typebox").TString;
-    urls: import("@sinclair/typebox").TOptional<import("@sinclair/typebox").TArray<import("@sinclair/typebox").TString>>;
-}>;
-export type AdvancedSearchInput = Static<typeof AdvancedSearchSchema>;
-export declare function advancedSearch(id: string, params: AdvancedSearchInput, signal: AbortSignal, onUpdate: AgentToolUpdateCallback | undefined, ctx: ExtensionContext): Promise<import("@mariozechner/pi-coding-agent").AgentToolResult<any>>;

package/dist/advanced_search.js

@@ -1,76 +0,0 @@
-import { Type } from "@sinclair/typebox";
-import { callApiStream, getConfig, applyCitations } from "./api.js";
-import { getModel, missingConfigResult, errorResult, formatResult } from "./utils.js";
-export const AdvancedSearchSchema = Type.Object({
-    query: Type.String({ description: "The search query or question to answer" }),
-    urls: Type.Optional(Type.Array(Type.String(), {
-        description: "Additional URLs to include in context (up to 20)",
-        maxItems: 20
-    })),
-});
-export async function advancedSearch(id, params, signal, onUpdate, ctx) {
-    const model = await getModel(ctx);
-    if (!model)
-        return missingConfigResult(ctx);
-    const hasUrls = params.urls && params.urls.length > 0;
-    const urlCount = hasUrls ? params.urls.length : 0;
-    onUpdate?.({
-        content: [{
-                type: "text",
-                text: hasUrls
-                    ? `Searching and analyzing ${urlCount} URL(s)...`
-                    : `Searching for "${params.query}"...`
-            }],
-        details: {}
-    });
-    try {
-        const config = getConfig(model);
-        // Build prompt: include URLs in the query if provided
-        const prompt = hasUrls
-            ? `${params.query}\n\nAlso analyze these URLs:\n${params.urls.join("\n")}`
-            : params.query;
-        // Enable both tools: google_search + url_context
-        // Gemini will decide when to use each
-        const tools = hasUrls
-            ? [{ [config.searchTool]: {} }, { [config.urlContextTool]: {} }]
-            : [{ [config.searchTool]: {} }];
-        const result = await callApiStream(ctx, model, {
-            contents: [{ role: "user", parts: [{ text: prompt }] }],
-            tools
-        }, onUpdate);
-        const { text, sources } = applyCitations(result.text, result.groundingMetadata);
-        // Handle URL context metadata
-        const urlMeta = result.urlContextMetadata?.urlMetadata
-            || result.urlContextMetadata?.url_metadata || [];
-        const retrieved = urlMeta
-            .filter((m) => (m.urlRetrievalStatus || m.url_retrieval_status) === "URL_RETRIEVAL_STATUS_SUCCESS")
-            .map((m) => m.retrievedUrl || m.retrieved_url || m.url);
-        const failed = urlMeta
-            .filter((m) => (m.urlRetrievalStatus || m.url_retrieval_status) !== "URL_RETRIEVAL_STATUS_SUCCESS")
-            .map((m) => ({
-            url: m.retrievedUrl || m.retrieved_url || m.url,
-            status: m.urlRetrievalStatus || m.url_retrieval_status
-        }));
-        let summary = text;
-        // Add URL status if there were failures
-        if (failed.length > 0) {
-            summary += `\n\n## URL Status\n✅ Retrieved: ${retrieved.length}\n❌ Failed: ${failed.length}`;
-            failed.forEach((f) => { summary += `\n- ${f.url}: ${f.status}`; });
-        }
-        // Add sources
-        if (sources.length > 0) {
-            summary += `\n\n## Sources\n${sources.map((s, i) => `${i + 1}. [${s.title}](${s.url})`).join("\n")}`;
-        }
-        return formatResult(summary, {
-            sources,
-            searchQueries: result.groundingMetadata?.webSearchQueries,
-            retrieved: retrieved.length > 0 ? retrieved : undefined,
-            failed: failed.length > 0 ? failed : undefined,
-            model: model.id,
-            grounded: sources.length > 0
-        });
-    }
-    catch (e) {
-        return errorResult(e);
-    }
-}

package/dist/youtube_video.d.ts

@@ -1,10 +0,0 @@
-import type { ExtensionContext, AgentToolUpdateCallback } from "@mariozechner/pi-coding-agent";
-import { type Static } from "@sinclair/typebox";
-export declare const YoutubeVideoSchema: import("@sinclair/typebox").TObject<{
-    video_url: import("@sinclair/typebox").TString;
-    query: import("@sinclair/typebox").TString;
-    start_offset: import("@sinclair/typebox").TOptional<import("@sinclair/typebox").TString>;
-    end_offset: import("@sinclair/typebox").TOptional<import("@sinclair/typebox").TString>;
-}>;
-export type YoutubeVideoInput = Static<typeof YoutubeVideoSchema>;
-export declare function youtubeVideo(id: string, params: YoutubeVideoInput, signal: AbortSignal, onUpdate: AgentToolUpdateCallback | undefined, ctx: ExtensionContext): Promise<import("@mariozechner/pi-coding-agent").AgentToolResult<any>>;

package/dist/youtube_video.js

@@ -1,54 +0,0 @@
-import { Type } from "@sinclair/typebox";
-import { callApiStream } from "./api.js";
-import { getModel, missingConfigResult, errorResult, formatResult } from "./utils.js";
-import { urlContext } from "./url_context.js";
-export const YoutubeVideoSchema = Type.Object({
-    video_url: Type.String({ description: "YouTube video URL" }),
-    query: Type.String({ description: "Question or task about the video" }),
-    start_offset: Type.Optional(Type.String({ description: "Start time (e.g., '120s' or '2:00')" })),
-    end_offset: Type.Optional(Type.String({ description: "End time (e.g., '300s' or '5:00')" })),
-});
-export async function youtubeVideo(id, params, signal, onUpdate, ctx) {
-    const model = await getModel(ctx);
-    if (!model)
-        return missingConfigResult(ctx);
-    // Only google-generative-ai supports native video; others use URL context
-    if (model.api !== "google-generative-ai") {
-        let query = params.query;
-        if (params.start_offset || params.end_offset) {
-            query += "\n\nFocus on the video section";
-            if (params.start_offset)
-                query += ` from ${params.start_offset}`;
-            if (params.end_offset)
-                query += ` to ${params.end_offset}`;
-            query += ".";
-        }
-        return urlContext(id, { query, urls: [params.video_url] }, signal, onUpdate, ctx);
-    }
-    onUpdate?.({ content: [{ type: "text", text: `Analyzing YouTube video...` }], details: {} });
-    try {
-        const videoPart = {
-            file_data: { file_uri: params.video_url, mime_type: "video/mp4" }
-        };
-        if (params.start_offset || params.end_offset) {
-            videoPart.video_metadata = {};
-            if (params.start_offset)
-                videoPart.video_metadata.start_offset = params.start_offset;
-            if (params.end_offset)
-                videoPart.video_metadata.end_offset = params.end_offset;
-        }
-        const result = await callApiStream(ctx, model, {
-            contents: [{ role: "user", parts: [videoPart, { text: params.query }] }]
-        }, onUpdate);
-        return formatResult(result.text, {
-            videoUrl: params.video_url,
-            clipping: (params.start_offset || params.end_offset)
-                ? { start: params.start_offset, end: params.end_offset }
-                : undefined,
-            model: model.id
-        });
-    }
-    catch (e) {
-        return errorResult(e);
-    }
-}

package/run.sh

@@ -1,9 +0,0 @@
-pi --no-skills --no-extensions -e ./src/index.ts --models "google-antigravity/*" '搜索今天btc价格'
-pi --no-skills --no-extensions -e ./src/index.ts --models "google-gemini-cli/*" '搜索今天btc价格'
-pi --no-skills --no-extensions -e ./src/index.ts --models "google/*" '搜索今天btc价格'
-
-pi --no-skills --no-extensions -e ./src/index.ts --provider google-antigravity --model claude-sonnet-4-5 '搜索今天btc价格'
-# ok
-pi --no-skills --no-extensions -e ./src/index.ts --provider google-gemini-cli --model gemini-3-flash-preview '搜索今天btc价格'
-# ok
-pi --no-skills --no-extensions -e ./src/index.ts --provider google --model gemini-2.5-flash '搜索今天btc价格'

package/src/youtube_video.ts

@@ -1,64 +0,0 @@
-import type { ExtensionContext, AgentToolUpdateCallback } from "@mariozechner/pi-coding-agent";
-import { Type, type Static } from "@sinclair/typebox";
-import { callApiStream } from "./api.js";
-import { getModel, missingConfigResult, errorResult, formatResult } from "./utils.js";
-import { urlContext } from "./url_context.js";
-
-export const YoutubeVideoSchema = Type.Object({
-    video_url: Type.String({ description: "YouTube video URL" }),
-    query: Type.String({ description: "Question or task about the video" }),
-    start_offset: Type.Optional(Type.String({ description: "Start time (e.g., '120s' or '2:00')" })),
-    end_offset: Type.Optional(Type.String({ description: "End time (e.g., '300s' or '5:00')" })),
-});
-export type YoutubeVideoInput = Static<typeof YoutubeVideoSchema>;
-
-export async function youtubeVideo(
-    id: string, 
-    params: YoutubeVideoInput, 
-    signal: AbortSignal, 
-    onUpdate: AgentToolUpdateCallback | undefined, 
-    ctx: ExtensionContext
-) {
-    const model = await getModel(ctx);
-    if (!model) return missingConfigResult(ctx);
-
-    // Only google-generative-ai supports native video; others use URL context
-    if (model.api !== "google-generative-ai") {
-        let query = params.query;
-        if (params.start_offset || params.end_offset) {
-            query += "\n\nFocus on the video section";
-            if (params.start_offset) query += ` from ${params.start_offset}`;
-            if (params.end_offset) query += ` to ${params.end_offset}`;
-            query += ".";
-        }
-        return urlContext(id, { query, urls: [params.video_url] }, signal, onUpdate, ctx);
-    }
-
-    onUpdate?.({ content: [{ type: "text", text: `Analyzing YouTube video...` }], details: {} });
-
-    try {
-        const videoPart: any = {
-            file_data: { file_uri: params.video_url, mime_type: "video/mp4" }
-        };
-
-        if (params.start_offset || params.end_offset) {
-            videoPart.video_metadata = {};
-            if (params.start_offset) videoPart.video_metadata.start_offset = params.start_offset;
-            if (params.end_offset) videoPart.video_metadata.end_offset = params.end_offset;
-        }
-
-        const result = await callApiStream(ctx, model, {
-            contents: [{ role: "user", parts: [videoPart, { text: params.query }] }]
-        }, onUpdate);
-
-        return formatResult(result.text, {
-            videoUrl: params.video_url,
-            clipping: (params.start_offset || params.end_offset) 
-                ? { start: params.start_offset, end: params.end_offset } 
-                : undefined,
-            model: model.id
-        });
-    } catch (e: any) {
-        return errorResult(e);
-    }
-}