Documentation

FAQ and troubleshooting

Start with the symptom below. Logs are available under the local LlamaBoss data folder when deeper diagnosis is needed.

Applies to LlamaBoss v0.1.10 · Updated 2026-07-24

• SmartScreen
• Model load
• Slow
• Vision
• Agent tools
• Remote
• Docs routes
• Logs

Windows says it protected your PC

When the installer came from llamaboss.com, choose More info → Run anyway. SmartScreen reputation is separate from traditional malware detection and is common for unsigned or newly signed independent applications.

A local model will not load

• Confirm the selected file is a valid GGUF—not an HTML download page or placeholder.
• Try a smaller model or close GPU-heavy apps.
• Return context length to 4K or 8K for the test.
• Enable 8-bit KV cache.
• Confirm the active models folder is the one that contains the file.
• Open the server log for the exact llama.cpp error.

Responses are much slower than expected

The model may be partially offloaded to system RAM, another window may currently be using the shared server, or the selected context may be consuming significant memory. Try a smaller model, a smaller context, and close other GPU workloads.

The model ignores an image

Use a vision-capable model and its correct projector. Text-only models cannot interpret images. If several unrelated mmproj files are present, move each vision model and its matching projector into a clean folder to remove ambiguity.

The model talks about a tool but does not use it

• Confirm the robot button is enabled.
• Try a model with stronger tool-calling behavior.
• For a remote endpoint, confirm the selected tool protocol matches the provider/model.
• Use a direct slash command to test the harness independently of the model, such as /pwd or /ls.
• Check whether the agent stopped at the tool-step cap or a loop guard.

A remote endpoint fails

• Verify Base URL and chat path.
• Verify Bearer versus X-API-Key authentication.
• Confirm the Endpoint provider/key names match an existing Connection.
• Use the exact wire model ID expected by the provider.
• Run /think auto if the endpoint rejects an optional reasoning field.

A documentation subpage shows the homepage

The deployment must include each nested docs/<guide>/index.html file. This documentation package also includes explicit Cloudflare Pages _redirects rewrites for every guide route. Deploy the contents of the website folder—not only the root-level files.

Where to look next

%LOCALAPPDATA%\LlamaBoss\logs\llamaboss.log
%LOCALAPPDATA%\LlamaBoss\logs\server.log

Remove secrets and personal paths before posting logs publicly. Report reproducible bugs through GitHub Issues.