Skip to content

Releases

agenthand v1.0.0-hackathon

Choose a tag to compare

@AUsernameForYou AUsernameForYou released this 04 Nov 22:59

agenthand v1.0.0-hackathon

agenthand stages product photos in new rooms and proves the product never changed, pixel by pixel.

Built for the Meta Global AI Developer Hackathon, October 26 to November 4, 2026. One MCP server, two independent builds (Rust and Python) that run the same Differential Conformance Suite.

Demo video

The demo film is attached to this release as a downloadable mp4 (the X copy, 1080p, 2:59). It is the same film the judges see: every staged photo in it carries a receipt, and the product pixels check out byte for byte.

Verify it yourself

Every claim in the Proof Graph can be re-checked with one command:

agenthand evidence verify --all

Licence

PolyForm Noncommercial 1.0.0. The source is public to read, run, and learn from. Commercial use needs a separate licence from the copyright holder.

agenthand 0.2.0 — Rust build

Choose a tag to compare

@AUsernameForYou AUsernameForYou released this 28 Oct 21:27

agenthand 0.2.0 — the Rust build becomes the public submission build.

One static binary per platform, no Python runtime. Same three tools
(identify, cut_out, draw), same receipts, measured parity with the
Python reference build: identical output pixel hash on the frozen
proof run (580,613 opaque pixels checked, 0 changed), identical
billed cost to the micro-dollar, ~540x faster cold start.

Assets

  • agenthand-0.2.0-<platform>.mcpb — one-click bundles for
    MCPB-supporting clients.
  • agenthand-0.2.0-<platform> — bare binaries for any MCP client
    config. Verify against SHA256SUMS.

Platform notes

  • macOS: binaries are unsigned and not notarized. Gatekeeper
    blocks the first launch of a browser-downloaded copy; right-click
    → Open, or xattr -d com.apple.quarantine <file>. Notarization is
    a planned, paid step.
  • Linux: the binary is a fully static musl build; it runs on any
    x86_64 distribution.
  • HEIC: the Rust build does not decode HEIC yet (a from-scratch
    decoder is in active development); it refuses HEIC loudly. The
    Python build handles HEIC today.