native tool calling detection for novita

server/utils/AiProviders/novita/index.js

@@ -82,7 +82,7 @@ class NovitaLLM {
   // from the current date. If it is, then we will refetch the API so that all the models are up
   // to date.
   #cacheIsStale() {
-    const MAX_STALE = 6.048e8; // 1 Week in MS
+    const MAX_STALE = 2.592e8; // 3 days in MS
     if (!fs.existsSync(this.cacheAtPath)) return true;
     const now = Number(new Date());
     const timestampMs = Number(fs.readFileSync(this.cacheAtPath));
@@ -143,6 +143,32 @@ class NovitaLLM {
     return availableModels[this.model]?.maxLength || 4096;
   }
 
+  /**
+   * Get the capabilities of a model from the Novita API.
+   * @returns {Promise<{tools: 'unknown' | boolean, reasoning: 'unknown' | boolean, imageGeneration: 'unknown' | boolean, vision: 'unknown' | boolean}>}
+   */
+  async getModelCapabilities() {
+    try {
+      await this.#syncModels();
+      const availableModels = this.models();
+      const modelInfo = availableModels[this.model];
+      return {
+        tools: modelInfo.features.includes("function-calling"),
+        reasoning: modelInfo.features.includes("reasoning"),
+        imageGeneration: false, // no image generation capabilities for Novita yet.
+        vision: modelInfo.input_modalities.includes("image"),
+      };
+    } catch (error) {
+      console.error("Error getting model capabilities:", error);
+      return {
+        tools: "unknown",
+        reasoning: "unknown",
+        imageGeneration: "unknown",
+        vision: "unknown",
+      };
+    }
+  }
+
   async isValidChatCompletionModel(model = "") {
     await this.#syncModels();
     const availableModels = this.models();
@@ -398,6 +424,8 @@ async function fetchNovitaModels() {
             model.id.split("/")[0].charAt(0).toUpperCase() +
             model.id.split("/")[0].slice(1),
           maxLength: model.context_size,
+          features: model.features ?? [],
+          input_modalities: model.input_modalities ?? [],
         };
       });
 

server/utils/agents/aibitat/providers/novita.js

@@ -2,9 +2,14 @@ const OpenAI = require("openai");
 const Provider = require("./ai-provider.js");
 const InheritMultiple = require("./helpers/classes.js");
 const UnTooled = require("./helpers/untooled.js");
+const { tooledStream, tooledComplete } = require("./helpers/tooled.js");
+const { RetryError } = require("../error.js");
+const { NovitaLLM } = require("../../../AiProviders/novita/index.js");
 
 /**
  * The agent provider for the Novita AI provider.
+ * Supports true OpenAI-compatible tool calling when the model supports it,
+ * falling back to the UnTooled prompt-based approach otherwise.
  */
 class NovitaProvider extends InheritMultiple([Provider, UnTooled]) {
   model;
@@ -25,8 +30,13 @@ class NovitaProvider extends InheritMultiple([Provider, UnTooled]) {
     this._client = client;
     this.model = model;
     this.verbose = true;
+    this._supportsToolCalling = null;
   }
 
+  /**
+   * Get the Novita client.
+   * @returns {import("openai").OpenAI}
+   */
   get client() {
     return this._client;
   }
@@ -36,12 +46,16 @@ class NovitaProvider extends InheritMultiple([Provider, UnTooled]) {
   }
 
   /**
-   * Whether this provider supports native OpenAI-compatible tool calling.
-   * Override in subclass and return true to use native tool calling instead of UnTooled.
-   * @returns {boolean|Promise<boolean>}
+   * Whether the loaded model supports native OpenAI-compatible tool calling.
+   * Checks the Novita model capabilities and caches the result.
+   * @returns {Promise<boolean>}
    */
-  supportsNativeToolCalling() {
-    return false;
+  async supportsNativeToolCalling() {
+    if (this._supportsToolCalling !== null) return this._supportsToolCalling;
+    const novita = new NovitaLLM(null, this.model);
+    const capabilities = await novita.getModelCapabilities();
+    this._supportsToolCalling = capabilities.tools === true;
+    return this._supportsToolCalling;
   }
 
   async #handleFunctionCallChat({ messages = [] }) {
@@ -70,33 +84,101 @@ class NovitaProvider extends InheritMultiple([Provider, UnTooled]) {
     });
   }
 
+  /**
+   * Stream a chat completion with tool calling support.
+   * Uses native tool calling when supported, otherwise falls back to UnTooled.
+   */
   async stream(messages, functions = [], eventHandler = null) {
-    return await UnTooled.prototype.stream.call(
-      this,
-      messages,
-      functions,
-      this.#handleFunctionCallStream.bind(this),
-      eventHandler
+    const useNative =
+      functions.length > 0 && (await this.supportsNativeToolCalling());
+
+    if (!useNative) {
+      return await UnTooled.prototype.stream.call(
+        this,
+        messages,
+        functions,
+        this.#handleFunctionCallStream.bind(this),
+        eventHandler
+      );
+    }
+
+    this.providerLog(
+      "Provider.stream (tooled) - will process this chat completion."
     );
+
+    try {
+      return await tooledStream(
+        this.client,
+        this.model,
+        messages,
+        functions,
+        eventHandler
+      );
+    } catch (error) {
+      console.error(error.message, error);
+      if (error instanceof OpenAI.AuthenticationError) throw error;
+      if (
+        error instanceof OpenAI.RateLimitError ||
+        error instanceof OpenAI.InternalServerError ||
+        error instanceof OpenAI.APIError
+      ) {
+        throw new RetryError(error.message);
+      }
+      throw error;
+    }
   }
 
+  /**
+   * Create a non-streaming completion with tool calling support.
+   * Uses native tool calling when supported, otherwise falls back to UnTooled.
+   */
   async complete(messages, functions = []) {
-    return await UnTooled.prototype.complete.call(
-      this,
-      messages,
-      functions,
-      this.#handleFunctionCallChat.bind(this)
-    );
+    const useNative =
+      functions.length > 0 && (await this.supportsNativeToolCalling());
+
+    if (!useNative) {
+      return await UnTooled.prototype.complete.call(
+        this,
+        messages,
+        functions,
+        this.#handleFunctionCallChat.bind(this)
+      );
+    }
+
+    try {
+      const result = await tooledComplete(
+        this.client,
+        this.model,
+        messages,
+        functions,
+        this.getCost.bind(this)
+      );
+
+      if (result.retryWithError) {
+        return this.complete([...messages, result.retryWithError], functions);
+      }
+
+      return result;
+    } catch (error) {
+      if (error instanceof OpenAI.AuthenticationError) throw error;
+      if (
+        error instanceof OpenAI.RateLimitError ||
+        error instanceof OpenAI.InternalServerError ||
+        error instanceof OpenAI.APIError
+      ) {
+        throw new RetryError(error.message);
+      }
+      throw error;
+    }
   }
 
   /**
    * Get the cost of the completion.
-   *
+   * Stubbed since Novita AI has no cost basis.
    * @param _usage The completion to get the cost for.
    * @returns The cost of the completion.
-   * Stubbed since Novita AI has no cost basis.
    */
-  getCost() {
+  getCost(_usage) {
     return 0;
   }
 }