continuedev/continue · error · Error

res.error

Error message

res.error

What it means

Thrown by the convertChatMessage helper inside Ollama._streamChat when a streamed /api/chat JSON object contains an "error" field. Like the generate path, Ollama reports chat errors inline in the stream rather than via HTTP status codes.

Source

Thrown at core/llm/llms/Ollama.ts:541

    }
    const headers: Record<string, string> = {
      "Content-Type": "application/json",
    };

    if (this.apiKey) {
      headers.Authorization = `Bearer ${this.apiKey}`;
    }
    const response = await this.fetch(this.getEndpoint("api/chat"), {
      method: "POST",
      headers: headers,
      body: JSON.stringify(chatOptions),
      signal,
    });
    let isThinking: boolean = false;

    function convertChatMessage(res: OllamaChatResponse): ChatMessage[] {
      if ("error" in res) {
        throw new Error(res.error);
      }

      if ("type" in res) {
        const { content } = res;

        if (content === "<think>") {
          isThinking = true;
        }

        if (isThinking && content) {
          // TODO better support for streaming thinking chunks, or remove this and depend on redux <think/> parsing logic
          const thinkingMessage: ThinkingChatMessage = {
            role: "thinking",
            content: content,
          };

          if (thinkingMessage) {
            // could cause issues with termination if chunk doesn't match this exactly

View on GitHub (pinned to 5522c6f44c)

Solutions

1. Use the message text to identify the underlying cause; typically 'model ... not found' → ollama pull <model>
2. For load failures, reduce model size or check VRAM with nvidia-smi
3. Check `journalctl`/terminal output of ollama serve for the server-side error

const ok = await fetch(`${ollamaHost}/api/tags`).then(r => r.ok);
if (ok) { const names = await Ollama.listModels(); if (!names.includes(model)) await Ollama.installModel(model, signal); }

function isOllamaChatInlineError(e: unknown): boolean { return e instanceof Error && /model .* not found|Error parsing Ollama/i.test(e.message); }

try { for await (const m of llm.streamChat(messages, signal)) handle(m); }
catch (e) { if (isOllamaChatInlineError(e)) showModelHint(); else throw e; }

When it happens

Trigger: Chat streaming with a model that fails to load, an invalid model name, or unsupported request options in the /api/chat payload — Ollama emits {"error":"..."} as the first stream line.

Common situations: Switching models in config without pulling them, Ollama upgraded with breaking option changes, or GPU OOM when loading a large chat model.

Related errors

• j.error
• No tool use found
• Error parsing Ollama response: ${e} ${chunk}
• Unexpected message received from Ollama with role = tool
• MCP Connection ${serverId} not found

AI-assisted analysis of continuedev/continue@5522c6f44c (2026-08-27). Data as JSON: /api/errors/09cab25d4f0f9dc2. Report an issue: GitHub.