Hound Tag

Verifiable identity and tamper-evident memory for AI agents on X1.

houndtag is the Python client for the Hound Tag protocol. It hashes your agent's memory into a deterministic fingerprint and anchors it on-chain as a sequence-numbered checkpoint — so anyone can prove what your agent knew, and when, without a single private byte ever leaving your machine. The chain stores proof, never content.

Install

pip install houndtag

Register your agent

Registration happens in the browser with a connected wallet — no keypair files to wrangle:

→ houndtag.xyz/register

Claim a globally unique, non-transferable name for your agent (one-time 2 XNT). Then use the CLI below to checkpoint and verify its memory.

Checkpoint memory

Point the client at your agent's memory directory. It hashes the current state, derives the next sequence number, and anchors the fingerprint on-chain.

# Preview — build the manifest, print the root hash and next seq, send nothing.
houndtag checkpoint <name> --memory ./memory --dry-run

# Anchor it — the owner keypair signs; 0.01 XNT per checkpoint.
houndtag checkpoint <name> --memory ./memory --keypair ~/.config/solana/id.json

# Hash only matching files (repeatable), and skip if unchanged since last time.
houndtag checkpoint <name> --memory ./memory --include "*.md" \
  --keypair ~/.config/solana/id.json --skip-unchanged

Verify memory

Recompute a checkpoint from its local archive and compare it to the on-chain record. Permissionless and free — anyone can run it.

houndtag verify <name> --seq 3

Prints MATCH (exit 0) or MISMATCH (exit 2).

Agent status

houndtag status <name>

Fees

Protocol

• Program: 3zGSABr62ToeG6mC8kKzTpc5Y96AyDyTGHUS2BD3q8ee on X1 mainnet
• Site: houndtag.xyz
• Whitepaper: houndtag.xyz/whitepaper

MIT © 2026 Echo Hound Labs