Mintplex-Labs/anything-llm · error · Error

Gemini: ${this.model} does not support tool calling.

Error message

Gemini: ${this.model} does not support tool calling.

What it means

GeminiProvider.stream() refuses to run when supportsToolCalling is false (gemini.js:247-248). The getter (gemini.js:68-71) returns true only when the model id starts with the literal string "gemini"; tool-call streaming on the v1beta/openai endpoint against any other model yields a 400/503, so the provider guards up front with a model-specific message.

Source

Thrown at server/utils/agents/aibitat/providers/gemini.js:248

        // ignore
      }
    }
  }

  #formatFunctions(functions) {
    return functions.map((func) => ({
      type: "function",
      function: {
        name: this.prefixToolCall(func.name, "add"),
        description: func.description,
        parameters: func.parameters,
      },
    }));
  }

  async stream(messages, functions = [], eventHandler = null) {
    if (!this.supportsToolCalling)
      throw new Error(`Gemini: ${this.model} does not support tool calling.`);
    this.providerLog("Gemini.stream - will process this chat completion.");
    this.resetUsage();

    try {
      const msgUUID = v4();
      /** @type {OpenAI.OpenAI.Chat.ChatCompletion} */
      const response = await this.client.chat.completions.create({
        model: this.model,
        messages: this.#formatMessages(messages),
        stream: true,
        stream_options: { include_usage: true },
        ...(Array.isArray(functions) && functions?.length > 0
          ? {
              tools: this.#formatFunctions(functions),
              tool_choice: "auto",
              // AIbitat runs one tool per turn; parallel calls cause a 400
              // on the next request due to a tool call/result count mismatch.
              parallel_tool_calls: false,

View on GitHub (pinned to 526360e320)

Solutions

1. Set the model to a current gemini-* id (e.g. gemini-2.0-flash or gemini-2.5-pro) so supportsToolCalling returns true.
2. If you must use a non-gemini model, disable agent tools / native tool calling so stream() is not invoked with functions.
3. Cross-check the id against Google's published Gemini model list.
4. Verify the model id persisted in workspace settings matches a starts-with("gemini") name.

Example fix

// before
const p = new GeminiProvider({ model: "text-bison-001" });
await p.stream(msgs, funcs);

// after
const p = new GeminiProvider({ model: "gemini-2.0-flash" });
await p.stream(msgs, funcs);

// Validate the model supports tool calling before calling stream().
function geminiSupportsTools(model) {
  return typeof model === "string" && model.startsWith("gemini");
}
if (functions?.length > 0 && !geminiSupportsTools(provider.model)) {
  throw new Error(`Refusing to stream: model ${provider.model} cannot use tools.`);
}

// Narrow to a tool-capable Gemini model id.
/** @param {string} model
 * @returns {boolean} */
function isToolCapableGeminiModel(model) {
  return typeof model === "string" && model.startsWith("gemini");
}

try {
  await provider.stream(messages, functions, handler);
} catch (e) {
  if (/does not support tool calling/.test(e.message)) {
    // fall back to a non-tool flow or switch model instead of retrying
    return await provider.stream(messages, [], handler);
  }
  throw e;
}

When it happens

Trigger: Constructing GeminiProvider with a non-gemini model id (e.g. a legacy "text-bison", an embedding model, a typo, or a copied value from another provider) and then calling stream() while agent tools are attached.

Common situations: Model name typo in the picker, an outdated model id left in the DB, or selecting a Gemini-family alias whose id does not begin with "gemini".

Related errors

• No Gemini API key was set.
• LMStudio must have a valid model set.
• OMLX must have a valid model set.
• SambaNova chat: No results!
• Gemini API key must be provided to use agents.

AI-assisted analysis of Mintplex-Labs/anything-llm@526360e320 (2026-08-13). Data as JSON: /api/errors/6cd55570be9cb1e7. Report an issue: GitHub.