Roadmap and FAQ

Roadmap

VersionStatusScope
0.2doneaudit HTML/MD/JSON, NVIDIA GPU/VRAM, disk, WSL, failure codes, hardened installer, proof pipeline, docs
0.3plannedlive-runtime integration suite (skipped without a model), llama.cpp completion timing, Ollama version details, AMD/Intel/Apple probes, KV-cache presets
0.4plannedconsent-based install planning with checksum and license gates, SBOM
0.5researchegress-test harness, signed audit capsules, local dashboard
1.0plannedindependent security review, per-IDE native activation matrix, externally reproduced demo

FAQ

Can my computer run a 7B model? Run localsutra fit. If it says possibly_fits, memory looks plausible. Load it in Ollama to find out for real.

Why does GPU say unknown? Only NVIDIA via nvidia-smi is implemented. Unknown means not determined.

Is my local AI private if the URL is 127.0.0.1? Not necessarily. The runtime may still use the internet. Test with a firewall or network monitor.

Does infer send my prompt anywhere? Only to your loopback Ollama. The report stores a SHA-256 of the prompt, not the text, and discards the output.

Why will compare not compare my two models? Different workloads or hardware are not comparable. The tool refuses instead of inventing a ranking.

Do skills run by themselves? No. They are instructions your coding agent may follow. The CLI works alone.

Is the audit a certification? No. It is a reproducible self-assessment with a not-tested list.

Can I trust a report I received? Run localsutra verify. A match proves the file is unchanged since sealing, not who made it.

Does it work offline? Yes. Everything except probe/infer needs no network at all, and those only use loopback.

How do I report a bug or vulnerability? Bugs: GitHub issues. Vulnerabilities: private advisory (see SECURITY.md).