K-Veritas
Documentation

K-Veritas CLI

K-Veritas ties a published result, or an AI agent's actions, to the exact code, hardware, and time that produced it. A hosted server signs each report but only ever sees hashes, never your data. Verifying needs no account and can run fully offline.

The tool does two things:

Experiment verification

Wrap a run to capture its metrics, provenance, and hardware evidence, then seal a signed PDF that anyone can verify.

Agent sessions

Record a signed, hash-chained log of everything an AI agent does, so no designated action can happen off the record.

Private by default. A sealed report reveals nothing sensitive: no code, filenames, data, weights, command line, or salt. You choose how much to disclose per run. See Provenance & privacy.

Installation

Install the CLI as a single static binary:

curl -fsSL https://github.com/27-GROUP/kveritas-releases/raw/main/bin/kveritas-linux-amd64 -o kveritas
chmod +x kveritas
sudo mv kveritas /usr/local/bin/

From source

If you have Go 1.22+ installed, build it yourself:

git clone https://github.com/27-GROUP/kveritas-go.git
cd kveritas-go
make build

Platforms

Prebuilt binaries cover Linux (amd64/arm64), macOS (Intel/Apple Silicon), and Windows. See the Download page. Some features are Linux-only; each guide says which, and the Reference has a platform table.

Quick start

From experiment to verified report in four commands:

kveritas init
kveritas run -- python train.py --epochs 50
kveritas seal --output report.pdf
kveritas verify report.pdf
  • initStart a session (redacted disclosure by default).
  • runWrap your experiment: capture metrics, provenance snapshots, and per-process hardware.
  • sealSign everything into a PDF report.
  • verifyLocal crypto checks plus the full server audit (add --offline for local-only).

Record a metric by printing a KVERITAS_METRIC line from your script, in any language. K-Veritas reads it from stdout:

print("KVERITAS_METRIC name=val_accuracy value=0.9471 step=100")

See Instrument your script for every directive, the field rules, and copy-paste snippets for PyTorch, Hugging Face, TensorFlow, and scikit-learn.

Reports can also be verified in the browser at kveritas.org/verify.