@saltcorn/large-language-model 0.9.9 → 0.9.11

package/constants.js

@@ -1,22 +1,26 @@
 const OPENAI_MODELS = [
-  "gpt-3.5-turbo",
-  "gpt-3.5-turbo-16k",
-  "gpt-4o-mini",
-  "gpt-4",
-  "gpt-4-32k",
-  "gpt-4-turbo-preview",
-  "gpt-4-turbo",
   "gpt-4o",
+  "gpt-4o-mini",
   "gpt-4.1",
   "gpt-4.1-mini",
   "gpt-4.1-nano",
-  "o1",
+  "gpt-5",
+  "gpt-5-mini",
+  "gpt-5-nano",
+  "gpt-5.1",
+  "gpt-5.1-mini",
+  "gpt-5.1-nano",
+  "gpt-5.2",
+  "gpt-5.2-pro",
   "o3",
   "o3-mini",
+  "o3-pro",
   "o4-mini",
-  "gpt-5",
-  "gpt-5-nano",
-  "gpt-5-mini",
+  "codex-mini-latest",
+  "gpt-5-codex",
+  "gpt-5.1-codex",
+  "gpt-5.1-codex-mini",
+  "gpt-5.1-codex-max",
 ];
 
 // https://github.com/ollama/ollama/blob/main/docs/faq.md#where-are-models-stored

package/generate.js

@@ -9,7 +9,9 @@ const {
 } = require("@google-cloud/aiplatform");
 const { google } = require("googleapis");
 const Plugin = require("@saltcorn/data/models/plugin");
+const File = require("@saltcorn/data/models/file");
 const path = require("path");
+const fs = require("fs");
 const { features, getState } = require("@saltcorn/data/db/state");
 const {
   generateText,
@@ -18,8 +20,10 @@ const {
   jsonSchema,
   embed,
   embedMany,
+  experimental_transcribe,
 } = require("ai");
 const { createOpenAI } = require("@ai-sdk/openai");
+const OpenAI = require("openai");
 let ollamaMod;
 if (features.esm_plugins) ollamaMod = require("ollama");
 
@@ -113,6 +117,67 @@ const getImageGeneration = async (config, opts) => {
   }
 };
 
+const getAudioTranscription = async (
+  { backend, apiKey, api_key, provider, ai_sdk_provider },
+  opts
+) => {
+  switch (backend) {
+    case "OpenAI":
+      const client = new OpenAI({
+        apiKey: opts?.api_key || api_key || apiKey,
+      });
+      const fp = opts.file.location
+        ? opts.file.location
+        : typeof opts.file === "string"
+        ? await (
+            await File.findOne(opts.file)
+          ).location
+        : null;
+      const model = opts?.model || "whisper-1";
+      const diarize = model === "gpt-4o-transcribe-diarize";
+      const transcript1 = await client.audio.transcriptions.create({
+        file: Buffer.isBuffer(opts.file) ? opts.file : fs.createReadStream(fp),
+
+        model,
+        ...(diarize
+          ? {
+              response_format: "diarized_json",
+              chunking_strategy: "auto",
+            }
+          : {}),
+      });
+      return transcript1;
+
+    case "AI SDK":
+      const api_Key = opts?.api_key || api_key || apiKey;
+      const prov_obj = createOpenAI({ apiKey: api_Key });
+      const audio =
+        opts.url ||
+        (Buffer.isBuffer(opts.file)
+          ? opts.file
+          : typeof opts.file === "string"
+          ? await (await File.findOne(opts.file)).get_contents()
+          : await opts.file.get_contents());
+      const extra = {};
+      if (opts.prompt)
+        extra.providerOptions = {
+          openai: {
+            prompt: opts.prompt,
+            //response_format: "text",
+          },
+        };
+      const transcript = await experimental_transcribe({
+        model: prov_obj.transcription(opts?.model || "whisper-1"),
+        audio,
+        ...extra,
+      });
+
+      return transcript;
+    default:
+      throw new Error("Audio transcription not implemented for this backend");
+  }
+};
+
 const getCompletion = async (config, opts) => {
   switch (config.backend) {
     case "AI SDK":
@@ -190,6 +255,16 @@ const getCompletion = async (config, opts) => {
   }
 };
 
+const getAiSdkModel = ({ provider, api_key, model_name }) => {
+  switch (provider) {
+    case "OpenAI":
+      const openai = createOpenAI({ apiKey: api_key });
+      return openai(model_name);
+    default:
+      throw new Error("Provider not found: " + provider);
+  }
+};
+
 const getCompletionAISDK = async (
   { apiKey, model, provider, temperature },
   {
@@ -204,13 +279,11 @@ const getCompletionAISDK = async (
   }
 ) => {
   const use_model_name = rest.model || model;
-  let model_obj;
-  switch (provider) {
-    case "OpenAI":
-      const openai = createOpenAI({ apiKey: api_key || apiKey });
-      model_obj = openai(use_model_name);
-      break;
-  }
+  let model_obj = getAiSdkModel({
+    model_name: use_model_name,
+    api_key: api_key || apiKey,
+    provider,
+  });
   const modifyChat = (chat) => {
     const f = (c) => {
       if (c.type === "image_url")
@@ -864,4 +937,9 @@ const getEmbeddingGoogleVertex = async (config, opts, oauth2Client) => {
   return embeddings;
 };
 
-module.exports = { getCompletion, getEmbedding, getImageGeneration };
+module.exports = {
+  getCompletion,
+  getEmbedding,
+  getImageGeneration,
+  getAudioTranscription,
+};

package/index.js

@@ -10,6 +10,7 @@ const {
   getCompletion,
   getEmbedding,
   getImageGeneration,
+  getAudioTranscription,
 } = require("./generate");
 const { OPENAI_MODELS } = require("./constants.js");
 const { eval_expression } = require("@saltcorn/data/models/expression");
@@ -400,6 +401,15 @@ const functions = (config) => {
       description: "Get vector embedding",
       arguments: [{ name: "prompt", type: "String" }],
     },
+    llm_transcribe: {
+      run: async (opts) => {
+        const result = await getAudioTranscription(config, opts);
+        return result;
+      },
+      isAsync: true,
+      description: "Get vector embedding",
+      arguments: [{ name: "prompt", type: "String" }],
+    },
   };
 };
 
@@ -640,6 +650,126 @@ module.exports = {
         else await table.updateRow(upd, row[table.pk_name]);
       },
     },
+    llm_transcribe_audio: {
+      description: "Generate text from audio file",
+      requireRow: true,
+      configFields: ({ table, mode }) => {
+        const override_fields =
+          config.backend === "OpenAI-compatible API" &&
+          (config.altconfigs || []).filter((c) => c.name).length
+            ? [
+                {
+                  name: "override_config",
+                  label: "Alternative LLM configuration",
+                  type: "String",
+                  attributes: { options: config.altconfigs.map((c) => c.name) },
+                },
+              ]
+            : [];
+
+        if (mode === "workflow") {
+          return [
+            {
+              name: "audio_file_field",
+              label: "Audio file variable",
+              sublabel: "Set the generated answer to this context variable",
+              type: "String",
+              required: true,
+            },
+            {
+              name: "answer_field",
+              label: "Response variable",
+              sublabel: "Set the generated response object to this context variable. The subfield <code>text</code> holds the string transcription",
+              type: "String",
+              required: true,
+            },
+            {
+              name: "model",
+              label: "The model name, for example <code>whisper-1</code>",
+              type: "String",
+            },
+            {
+              name: "prompt_template",
+              label: "Prompt",
+              sublabel:
+                "Additional prompt text (only some models). Use interpolations {{ }} to access variables in the context",
+              type: "String",
+              fieldview: "textarea",
+            },
+
+            //...override_fields,
+          ];
+        } else if (table) {
+          const textFields = table.fields
+            .filter((f) => f.type?.sql_name === "text")
+            .map((f) => f.name);
+          const fileFields = table.fields
+            .filter((f) => f.type === "File")
+            .map((f) => f.name);
+
+          return [
+            {
+              name: "audio_file_field",
+              label: "Audio file variable",
+              sublabel: "Set the generated answer to this context variable",
+              type: "String",
+              required: true,
+              attributes: { options: fileFields },
+            },
+            {
+              name: "answer_field",
+              label: "Answer field",
+              sublabel: "Output field will be set to the generated answer",
+              type: "String",
+              required: true,
+              attributes: { options: textFields },
+            },
+            {
+              name: "model",
+              label: "The model name, for example <code>whisper-1</code>",
+              type: "String",
+            },
+            {
+              name: "prompt_template",
+              label: "Prompt",
+              sublabel:
+                "Additional prompt text (only some models). Use interpolations {{ }} to access variables in the row",
+              type: "String",
+              fieldview: "textarea",
+            },
+            //...override_fields,
+          ];
+        }
+      },
+      run: async ({
+        row,
+        table,
+        user,
+        mode,
+        configuration: {
+          audio_file_field,
+          prompt_template,
+          answer_field,
+          //override_config,
+          model,
+        },
+      }) => {
+        const opts = { file: row[audio_file_field] };
+        if (prompt_template)
+          opts.prompt = interpolate(prompt_template, row, user);
+
+        if (model) opts.model = model;
+
+        const ans = await getAudioTranscription(config, opts);
+
+        if (mode === "workflow") return { [answer_field]: ans };
+        else
+          await table.updateRow(
+            { [answer_field]: ans.text },
+            row[table.pk_name]
+          );
+      },
+    },
     llm_generate_image: {
       description: "Generate image with AI based on a text prompt",
       requireRow: true,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@saltcorn/large-language-model",
-  "version": "0.9.9",
+  "version": "0.9.11",
   "description": "Large language models and functionality for Saltcorn",
   "main": "index.js",
   "dependencies": {
@@ -12,7 +12,8 @@
     "@google-cloud/aiplatform": "^3.34.0",
     "googleapis": "^144.0.0",
     "ai": "5.0.44",
-    "@ai-sdk/openai": "2.0.30"
+    "@ai-sdk/openai": "2.0.30",
+    "openai": "6.16.0"
   },
   "author": "Tom Nielsen",
   "license": "MIT",