Your AI agent did something. Prove exactly what, and when.

Self-reported logs are not evidence

An autonomous agent that sends messages, changes systems, or moves value produces a log — a text file its operator controls. A log that can be edited after the fact is indistinguishable from one written at the time. When a question arrives — a client disputing what was delivered, a counterparty disputing what was sent, your own postmortem of an agent gone sideways — the log's value depends on being able to show this is what the record said then, and it has not changed since. A file on the operator's disk cannot show that.

What anchoring adds

Orphograph gives each consequential action record a Bitcoin-anchored receipt: the record is hashed on your machine, the fingerprint is anchored to the Bitcoin blockchain, and you get a dated receipt plus a hosted certificate page anyone you authorize can check in their own browser. Three properties do the work:

• Dated. The record provably existed by the time of a specific Bitcoin block. A receipt anchored seconds after the action cannot have been written after the dispute started.
• Tamper-evident. Change one byte of the record — a recipient, an amount, a timestamp — and it no longer matches the receipt.
• Verifiable without trusting the operator. The proof is an open format checked against the public chain; the verifier is open source. The person relying on the record needs no account and no relationship with the person who made it.

If you are the one accepting agent work

The receipt is built for your side of the table. Before you rely on a delivered artifact or a claimed action, ask for its receipt and verify it yourself — drop the file or record you were given onto the acceptance page and check it against the certificate. A "yes" means the bytes in front of you are the bytes that existed at the anchored time. Operators who anchor as they work can answer any dispute with a link; operators who cannot produce receipts are asking you to take their logs on faith.

What this is not

Honesty is the product, so the scope is narrow and stated plainly. A receipt is not proof that an action was correct, authorized, safe, or legal, and anchoring is not a compliance regime or an agent-safety mechanism. It makes the record of what happened tamper-evident and independently datable — which is what turns a log into evidence. What that evidence establishes in any particular dispute is a question for that forum.

Run it inside the agent

A dependency-free CLI plus a drop-in skill let the agent anchor its own trail as it works — one-line action records, artifacts it produces, and a daily manifest of its memory files:

# Anchor a one-line action record the moment it happens
echo "sent invoice #42 to client-7" | python3 orpho_agent_anchor.py anchor-text --label "billing"

# Anchor an artifact the agent produced
python3 orpho_agent_anchor.py anchor-file ./deliverable.zip --label "handoff"

# Nightly: anchor a manifest of the agent's memory files
python3 orpho_agent_anchor.py anchor-memory ./workspace --label "memory"

MCP-capable agents can anchor through the MCP server instead — proof of existence becomes part of the agent's output, not an afterthought. Setup for both lives on the integrations page.

FAQ