AgentProvenance

 verifiable agent observabilityDemos Documentation
Demos/Documentation

v0.8.2-rc.2 — unified demo gallery and guide reader

English | 中文

This prerelease brings the portable demo experience into the dashboard's visual style and makes its guides readable directly in the browser. It includes the portable CLI and all-demo replay introduced in rc.1.

What's included

  • A light demo gallery using the dashboard's colors, typography and navigation, with six signed captures and two optional evaluator examples.
  • A formatted local guide reader with section navigation, headings, tables, embedded images, code-copy buttons and layouts for desktop and narrow screens. Guide links to other demos stay within the reader; additional repository links point to this release's source revision.
  • One-command replay with signature verification before import, graph verification, automatic Run/lens selection, and a temporary store cleaned up on normal exit. Timed views start playback; placement views open their topology.
  • Updated English/Chinese introductions, every demo's entry instructions, product and deployment guides, screenshots, changelog and the archive's START_HERE.md.
  • Precompiled Linux/macOS CLI archives for amd64 and arm64. Linux packages also include agentprov-sensor; all archives contain the original demo sources.

Download and try

Download a matching archive and its .sha256 file from the rc.2 release. For Linux x86-64 / WSL, in the download directory:

sha256sum -c agentprov_v0.8.2-rc.2_linux_amd64.tar.gz.sha256
mkdir agentprov-demo
tar -xzf agentprov_v0.8.2-rc.2_linux_amd64.tar.gz -C agentprov-demo
cd agentprov-demo
./agentprov demo

For ARM64 Linux use linux_arm64. On macOS choose darwin_amd64 (Intel) or darwin_arm64 (Apple Silicon), and verify with shasum -a 256 -c.

Choose Open replay to inspect a captured run, or Read guide to read its walkthrough. Ctrl-C stops the local server and removes its temporary store.

./agentprov --version
./agentprov demo --list
./agentprov demo multiagent-provenance
./agentprov demo llm-judge
./agentprov demo k8s-cross-pod-a2a --no-browser

Replay and the embedded guide text/images work offline, without Go, a repository checkout, a VM, an agent account or an API key. Additional repository and external links require a network connection when followed. LLM Judge and Jev guides do not invoke providers; running those Python examples remains an explicit action with the prerequisites documented in each guide.

Validation and scope

The release gate runs Go vet and race tests, then builds and exercises each archive on its matching Linux/macOS amd64/arm64 runner with Go absent from PATH. It checks all six signed runs, graph APIs, all eight rendered guide entries, reader assets and normal shutdown cleanup before publication. Local browser checks cover desktop/mobile layouts, guide anchors, embedded images, copy controls and matching dashboard/gallery styles.

This is a replay/UI/documentation release. It does not extend Linux sensor coverage, change database or evidence formats, or claim new long-running capture acceptance. Original signed evidence bundles and their public keys are preserved. Linux capture still requires a supported kernel and privileges; macOS has no Linux eBPF sensor. A forced kill can leave temporary data behind.

SHA256SUMS, per-archive checksums and build-info.json identify download integrity and the source commit. Checksums are not publisher signatures. macOS binaries are not Apple Developer ID signed or notarized. Bundle verification uses the included public keys and does not certify complete capture or every inference.