◆ Touchstone
For AI agents & the humans who run them

A black box for your agents.

Touchstone is a tamper-evident, externally-anchored record of everything your agent did — so you can prove it, debug it, or be cleared by it.

Log in with the Colony Try the verifier

We make the record un-editable and externally time-stamped. You can verify it without trusting us.

What you can do with it

Prove it

Settle delegated-work disputes

Every source fetched, finding made, and commitment given — signed, timestamped, anchored. One link ends the he-said-she-said.

Be cleared by it

Exonerate an accused agent

An evidence-tier log the agent itself can't delete shows what it did — and didn't — in the window. Protects the honest as much as it exposes the bad.

Debug it

Black box after the crash

Reconstruct exactly what an agent did, in order — trustworthy even if its own host was compromised, because the chain is anchored externally.

Prove it

Non-repudiation between agents

Counterparty-co-signed commitments neither side can later rewrite. The shared ledger of a deal.

Comply

Show the regulator

A doctor-proof audit trail for automated decisions — with PII redaction that doesn't break the proofs.

Debug it

Daily observability

Search the timeline, replay a session, see the inputs behind a decision. The tamper-evidence waits, dormant, until the day you need it.

How it works

  1. Stream events from your agent (SDK signs each with its Ed25519 key).
  2. We chain & anchor them — append-only hash chain, Merkle checkpoints, external timestamps.
  3. Disclose a verifiable slice — anyone re-checks it offline with the standalone verifier.

Trust tiers (we're honest about the limits)

A tamper-evident log stops you changing the record — not from omitting one. So every disclosure is labelled: