Verdict
This usually points to model size, offload path, memory bandwidth, thermals, or context length. Start with the simplest visible checks before changing advanced settings.
Diagnostic overview
This usually points to model size, offload path, memory bandwidth, thermals, or context length. Start with the simplest visible checks before changing advanced settings.
Stop troubleshooting and switch paths when the workflow involves confidential documents and you cannot explain where prompts, files, embeddings, logs, and provider calls go.
Symptoms
- - The local AI workflow does not behave as expected.
- - Local AI Guide has not independently reproduced this diagnostic.
Likely causes
- - Model size, CPU/offload path, context length, or hardware limits may be the bottleneck.
First fixes
- - Confirm the exact app, model, route, and error text.
- - Check whether the selected model is realistic for the machine’s RAM, VRAM, and context length.
- - Confirm whether the issue is a local runtime problem, a UI/provider problem, or a document/retrieval problem.
Stop and switch when
- - The workflow involves confidential data and the local/private boundary is unclear.
- - The hardware cannot realistically fit the selected model.
- - Stop troubleshooting and switch paths when the workflow involves confidential documents and you cannot explain where prompts, files, embeddings, logs, and provider calls go.