Mintplex-Labs · timothycarambat · Jan 24, 2025 · Jan 24, 2025 · Jan 24, 2025 · Jan 24, 2025
diff --git a/frontend/src/components/LLMSelection/TogetherAiOptions/index.jsx b/frontend/src/components/LLMSelection/TogetherAiOptions/index.jsx
@@ -2,6 +2,9 @@ import System from "@/models/system";
 import { useState, useEffect } from "react";
 
 export default function TogetherAiOptions({ settings }) {
+  const [inputValue, setInputValue] = useState(settings?.TogetherAiApiKey);
+  const [apiKey, setApiKey] = useState(settings?.TogetherAiApiKey);
+
   return (
     <div className="flex gap-[36px] mt-1.5">
       <div className="flex flex-col w-60">
@@ -17,37 +20,49 @@ export default function TogetherAiOptions({ settings }) {
           required={true}
           autoComplete="off"
           spellCheck={false}
+          onChange={(e) => setInputValue(e.target.value)}
+          onBlur={() => setApiKey(inputValue)}
         />
       </div>
       {!settings?.credentialsOnly && (
-        <TogetherAiModelSelection settings={settings} />
+        <TogetherAiModelSelection settings={settings} apiKey={apiKey} />
       )}
     </div>
   );
 }
-function TogetherAiModelSelection({ settings }) {
+
+function TogetherAiModelSelection({ settings, apiKey }) {
   const [groupedModels, setGroupedModels] = useState({});
   const [loading, setLoading] = useState(true);
 
   useEffect(() => {
     async function findCustomModels() {
       setLoading(true);
-      const { models } = await System.customModels("togetherai");
-
-      if (models?.length > 0) {
-        const modelsByOrganization = models.reduce((acc, model) => {
-          acc[model.organization] = acc[model.organization] || [];
-          acc[model.organization].push(model);
-          return acc;
-        }, {});
-
-        setGroupedModels(modelsByOrganization);
+      try {
+        const key = apiKey === "*".repeat(20) ? null : apiKey;
+        const { models } = await System.customModels("togetherai", key);
+        if (models?.length > 0) {
+          const modelsByOrganization = models.reduce((acc, model) => {
+            if (model.type !== "chat") return acc; // Only show chat models in dropdown
+            const org = model.organization || "Unknown";
+            acc[org] = acc[org] || [];
+            acc[org].push({
+              id: model.id,
+              name: model.name || model.id,
+              organization: org,
+              maxLength: model.maxLength,
+            });
+            return acc;
+          }, {});
+          setGroupedModels(modelsByOrganization);
+        }
+      } catch (error) {
+        console.error("Error fetching Together AI models:", error);
       }
-
       setLoading(false);
     }
     findCustomModels();
-  }, []);
+  }, [apiKey]);
 
   if (loading || Object.keys(groupedModels).length === 0) {
     return (

diff --git a/server/storage/models/.gitignore b/server/storage/models/.gitignore
@@ -5,4 +5,5 @@ openrouter
 apipie
 novita
 mixedbread-ai*
-gemini
+gemini
+togetherAi
diff --git a/server/utils/AiProviders/togetherAi/index.js b/server/utils/AiProviders/togetherAi/index.js
@@ -5,10 +5,76 @@ const {
 const {
   LLMPerformanceMonitor,
 } = require("../../helpers/chat/LLMPerformanceMonitor");
+const fs = require("fs");
+const path = require("path");
+const { safeJsonParse } = require("../../http");
 
-function togetherAiModels() {
-  const { MODELS } = require("./models.js");
-  return MODELS || {};
+const cacheFolder = path.resolve(
+  process.env.STORAGE_DIR
+    ? path.resolve(process.env.STORAGE_DIR, "models", "togetherAi")
+    : path.resolve(__dirname, `../../../storage/models/togetherAi`)
+);
+
+async function togetherAiModels(apiKey = null) {
+  const cacheModelPath = path.resolve(cacheFolder, "models.json");
+  const cacheAtPath = path.resolve(cacheFolder, ".cached_at");
+
+  // If cache exists and is less than 1 week old, use it
+  if (fs.existsSync(cacheModelPath) && fs.existsSync(cacheAtPath)) {
+    const now = Number(new Date());
+    const timestampMs = Number(fs.readFileSync(cacheAtPath));
+    if (now - timestampMs <= 6.048e8) {
+      // 1 Week in MS
+      return safeJsonParse(
+        fs.readFileSync(cacheModelPath, { encoding: "utf-8" }),
+        []
+      );
+    }
+  }
+
+  try {
+    const { OpenAI: OpenAIApi } = require("openai");
+    const openai = new OpenAIApi({
+      baseURL: "https://api.together.xyz/v1",
+      apiKey: apiKey || process.env.TOGETHER_AI_API_KEY || null,
+    });
+
+    const response = await openai.models.list();
+
+    // Filter and transform models into the expected format
+    // Only include chat models
+    const validModels = response.body
+      .filter((model) => ["chat"].includes(model.type))
+      .map((model) => ({
+        id: model.id,
+        name: model.display_name || model.id,
+        organization: model.organization || "Unknown",
+        type: model.type,
+        maxLength: model.context_length || 4096,
+      }));
+
+    // Cache the results
+    if (!fs.existsSync(cacheFolder))
+      fs.mkdirSync(cacheFolder, { recursive: true });
+    fs.writeFileSync(cacheModelPath, JSON.stringify(validModels), {
+      encoding: "utf-8",
+    });
+    fs.writeFileSync(cacheAtPath, String(Number(new Date())), {
+      encoding: "utf-8",
+    });
+
+    return validModels;
+  } catch (error) {
+    console.error("Error fetching Together AI models:", error);
+    // If cache exists but is stale, still use it as fallback
+    if (fs.existsSync(cacheModelPath)) {
+      return safeJsonParse(
+        fs.readFileSync(cacheModelPath, { encoding: "utf-8" }),
+        []
+      );
+    }
+    return [];
+  }
 }
 
 class TogetherAiLLM {
@@ -60,29 +126,34 @@ class TogetherAiLLM {
     return content.flat();
   }
 
-  allModelInformation() {
-    return togetherAiModels();
+  async allModelInformation() {
+    const models = await togetherAiModels();
+    return models.reduce((acc, model) => {
+      acc[model.id] = model;
+      return acc;
+    }, {});
   }
 
   streamingEnabled() {
     return "streamGetChatCompletion" in this;
   }
 
-  static promptWindowLimit(modelName) {
-    const availableModels = togetherAiModels();
-    return availableModels[modelName]?.maxLength || 4096;
+  static async promptWindowLimit(modelName) {
+    const models = await togetherAiModels();
+    const model = models.find((m) => m.id === modelName);
+    return model?.maxLength || 4096;
   }
 
-  // Ensure the user set a value for the token limit
-  // and if undefined - assume 4096 window.
-  promptWindowLimit() {
-    const availableModels = this.allModelInformation();
-    return availableModels[this.model]?.maxLength || 4096;
+  async promptWindowLimit() {
+    const models = await togetherAiModels();
+    const model = models.find((m) => m.id === this.model);
+    return model?.maxLength || 4096;
   }
 
   async isValidChatCompletionModel(model = "") {
-    const availableModels = this.allModelInformation();
-    return availableModels.hasOwnProperty(model);
+    const models = await togetherAiModels();
+    const foundModel = models.find((m) => m.id === model);
+    return foundModel && foundModel.type === "chat";
   }
 
   constructPrompt({
-Original file line number
+Diff line change
@@ Expand Up / @@ -5,4 +5,5 @@ openrouter @@
     apipie
     novita
     mixedbread-ai*
-    gemini
+    gemini
+    togetherAi