@zhexin/pi-deepseek 0.0.2 → 0.1.0

package/README.md

@@ -2,7 +2,7 @@
 
 DeepSeek Enhancement Extension for the [Pi](https://pi.dev) coding agent.
 
-Include **web search/fetch** tools. Nothing more.
+Include **web search/fetch**, **OCR**, and **ASR** tools. Nothing more.
 
 Most extensions I've tried seem to be compensating for something. Config that fights pi's philosophy, feature lists that look like a spaceship's dashboard, and UIs that assume you wanted a personal assistant.
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@zhexin/pi-deepseek",
   "description": "DeepSeek Enhancement for Pi",
-  "version": "0.0.2",
+  "version": "0.1.0",
   "type": "module",
   "license": "MIT",
   "keywords": [

package/src/index.ts

@@ -1,10 +1,22 @@
-import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
+import { openAsBlob } from "node:fs";
+import { basename, resolve } from "node:path";
+
+import type {
+  ExtensionAPI,
+  ExtensionContext,
+  ToolNamespace,
+} from "@earendil-works/pi-coding-agent";
 import type { Static } from "typebox";
 
-import { fetchInput, fetchOutput, searchInput, searchOutput } from "./schema.ts";
+import * as schema from "./schema.ts";
 
 const unauthorized = new Error("Unauthorized. Run `/login deepseek` to authenticate.");
 
+const deepseekNamespace: ToolNamespace = {
+  name: "deepseek",
+  description: "DeepSeek web search/fetch, document/image OCR, and speech-to-text ASR tools",
+};
+
 async function resolveDeepSeekRequest(ctx: ExtensionContext, path: string) {
   const authResult = await ctx.modelRegistry.getProviderAuth("deepseek");
   const apiKey = authResult?.auth.apiKey;
@@ -20,7 +32,6 @@ async function resolveDeepSeekRequest(ctx: ExtensionContext, path: string) {
     headers: {
       ...authResult?.auth.headers,
       Authorization: `Bearer ${apiKey}`,
-      "Content-Type": "application/json",
     },
   };
 }
@@ -29,15 +40,17 @@ export default function (pi: ExtensionAPI): void {
   pi.registerTool({
     name: "web-search",
     label: "DeepSeek Web Search",
+    namespace: deepseekNamespace,
     description: "Search web pages for a query and return structured results.",
-    parameters: searchInput,
-    outputSchema: searchOutput,
+    promptSnippet: "Search the web for current information",
+    parameters: schema.searchInput,
+    outputSchema: schema.searchOutput,
     async execute(_toolCallId, params, signal, _onUpdate, ctx) {
       const { url, headers } = await resolveDeepSeekRequest(ctx, "/v1/tools/search");
 
       const response = await fetch(url, {
         method: "POST",
-        headers,
+        headers: { ...headers, "Content-Type": "application/json" },
         body: JSON.stringify(params),
         signal: signal ?? null,
       });
@@ -51,7 +64,7 @@ export default function (pi: ExtensionAPI): void {
         );
       }
 
-      const data = (await response.json()) as Static<typeof searchOutput>;
+      const data = (await response.json()) as Static<typeof schema.searchOutput>;
       const formatted = data.results
         .map((r, i) =>
           [
@@ -75,15 +88,17 @@ export default function (pi: ExtensionAPI): void {
   pi.registerTool({
     name: "web-fetch",
     label: "DeepSeek Web Fetch",
+    namespace: deepseekNamespace,
     description: "Fetch a single web page and return its content as Markdown.",
-    parameters: fetchInput,
-    outputSchema: fetchOutput,
+    promptSnippet: "Fetch a web page by URL and return its content as Markdown",
+    parameters: schema.fetchInput,
+    outputSchema: schema.fetchOutput,
     async execute(_toolCallId, params, signal, _onUpdate, ctx) {
       const { url, headers } = await resolveDeepSeekRequest(ctx, "/v1/tools/fetch");
 
       const response = await fetch(url, {
         method: "POST",
-        headers,
+        headers: { ...headers, "Content-Type": "application/json" },
         body: JSON.stringify(params),
         signal: signal ?? null,
       });
@@ -97,7 +112,7 @@ export default function (pi: ExtensionAPI): void {
         );
       }
 
-      const data = (await response.json()) as Static<typeof fetchOutput>;
+      const data = (await response.json()) as Static<typeof schema.fetchOutput>;
       return {
         content: [{ type: "text", text: `Title: ${data.title}\n\n${data.content}` }],
         structuredContent: data,
@@ -105,4 +120,107 @@ export default function (pi: ExtensionAPI): void {
       };
     },
   });
+
+  pi.registerTool({
+    name: "ocr",
+    label: "DeepSeek OCR",
+    namespace: deepseekNamespace,
+    description:
+      "Recognize text in a PDF, Word (DOC, DOCX), PowerPoint (PPT, PPTX), or image (PNG, JPEG, WebP, GIF, BMP, TIFF) file page by page.",
+    promptSnippet: "Extract text from PDF, Word, PowerPoint, or image files via OCR",
+    promptGuidelines: [
+      "Use ocr instead of read for .pdf, .doc, .docx, .ppt, and .pptx files, or when extracting dense text from document images.",
+    ],
+    parameters: schema.ocrInput,
+    outputSchema: schema.ocrOutput,
+    async execute(_toolCallId, params, signal, _onUpdate, ctx) {
+      const { url, headers } = await resolveDeepSeekRequest(ctx, "/v1/tools/ocr");
+      const filePath = resolve(ctx.cwd, params.path.replace(/^@/, ""));
+      const blob = await openAsBlob(filePath);
+
+      const formData = new FormData();
+      formData.append("file", blob, basename(filePath));
+      formData.append("purpose", params.purpose);
+
+      const response = await fetch(url, {
+        method: "POST",
+        headers,
+        body: formData,
+        signal: signal ?? null,
+      });
+
+      if (!response.ok) {
+        if (response.status === 401) throw unauthorized;
+
+        const errorText = await response.text().catch(() => "");
+        throw new Error(
+          `DeepSeek OCR API error (status ${response.status}): ${errorText || response.statusText}`,
+        );
+      }
+
+      const data = (await response.json()) as
+        | Static<typeof schema.ocrOutput>
+        | { error: { message?: string; code?: string } };
+
+      if ("error" in data) {
+        throw new Error(
+          `DeepSeek OCR API error: ${data.error.message || data.error.code || "Recognition failed"}`,
+        );
+      }
+
+      const formatted = data.pages
+        .map((p) => `--- Page ${p.index + 1} ---\n${p.text}`)
+        .join("\n\n");
+
+      return {
+        content: [{ type: "text", text: formatted || "No text recognized." }],
+        structuredContent: data,
+        details: {},
+      };
+    },
+  });
+
+  pi.registerTool({
+    name: "asr",
+    label: "DeepSeek ASR",
+    exposure: "codemode",
+    namespace: deepseekNamespace,
+    description:
+      "Transcribe speech from an audio file (16-bit PCM WAV or Ogg Opus, max 8 MiB, max 120 seconds) into text.",
+    promptSnippet: "Transcribe speech from WAV or Ogg Opus audio files into text",
+    promptGuidelines: ["Use asr to transcribe speech from .wav or .ogg/.opus audio files."],
+    parameters: schema.asrInput,
+    outputSchema: schema.asrOutput,
+    async execute(_toolCallId, params, signal, _onUpdate, ctx) {
+      const { url, headers } = await resolveDeepSeekRequest(ctx, "/v1/tools/asr");
+      const filePath = resolve(ctx.cwd, params.path.replace(/^@/, ""));
+      const blob = await openAsBlob(filePath);
+
+      const formData = new FormData();
+      formData.append("file", blob, basename(filePath));
+
+      const response = await fetch(url, {
+        method: "POST",
+        headers,
+        body: formData,
+        signal: signal ?? null,
+      });
+
+      if (!response.ok) {
+        if (response.status === 401) throw unauthorized;
+
+        const errorText = await response.text().catch(() => "");
+        throw new Error(
+          `DeepSeek ASR API error (status ${response.status}): ${errorText || response.statusText}`,
+        );
+      }
+
+      const data = (await response.json()) as Static<typeof schema.asrOutput>;
+      return {
+        content: [{ type: "text", text: data.text }],
+        structuredContent: data,
+        details: {},
+      };
+    },
+  });
 }

package/src/schema.ts

@@ -63,3 +63,49 @@ export const fetchOutput = Type.Object({
   title: Type.String(),
   content: Type.String(),
 });
+
+export const ocrInput = Type.Object({
+  path: Type.String({
+    minLength: 1,
+    description:
+      "Path to the file to recognize. Supported: PDF, Word (DOC, DOCX), PowerPoint (PPT, PPTX), and images (PNG, JPEG, WebP, GIF, BMP, TIFF).",
+  }),
+  purpose: Type.Enum(["text_recognition", "rich_layout_recognition"], {
+    description:
+      "text_recognition extracts text directly from Word and PowerPoint files without images and recognizes other files page by page. rich_layout_recognition recognizes every file page by page.",
+  }),
+});
+
+export const ocrOutput = Type.Object({
+  id: Type.String(),
+  object: Type.Literal("ocr.result"),
+  created: Type.Integer(),
+  purpose: Type.Enum(["text_recognition", "rich_layout_recognition"]),
+  page_count: Type.Integer(),
+  pages: Type.Array(
+    Type.Object({
+      index: Type.Integer(),
+      text: Type.String(),
+      finish_reason: Type.Enum(["stop", "length"]),
+    }),
+  ),
+  usage: Type.Object({
+    input_tokens: Type.Integer(),
+    output_tokens: Type.Integer(),
+    total_tokens: Type.Integer(),
+  }),
+});
+
+export const asrInput = Type.Object({
+  path: Type.String({
+    minLength: 1,
+    description:
+      "Path to the audio file to transcribe (16-bit PCM WAV or Ogg Opus, max 8 MiB, max 120 seconds).",
+  }),
+});
+
+export const asrOutput = Type.Object({
+  request_id: Type.String(),
+  text: Type.String(),
+  duration_ms: Type.Integer(),
+});