Seal where the work happens
Anchoring one file by hand settles one argument. The stronger habit is a workspace that requests receipts as part of its normal workflow. A receipt exists only when the integration reports a successful anchor; confirmation against Bitcoin may occur later. These integrations hash file bytes locally. Single-file integrations submit digests; folder and dataset flows may also submit manifest paths, sizes, and digests.
Folders
Watch-folder sealer
A single-file, dependency-free Python daemon. Point it at a delivery folder; anything new or changed is fingerprinted and anchored once it stops changing. Receipts accumulate in a ledger inside the folder.
curl -O https://orphograph.com/dist/orphograph-watch.py python3 orphograph-watch.py ~/DeliveriesDownload orphograph-watch.py →
Releases
GitHub Action
Request one receipt per matched build artifact. Successful rows are
written to the job summary and orphograph-receipts.json in the
workspace. The default is fail_on_error: false, so the job
continues when anchoring is incomplete and the summary lists what was not
anchored. Set it to true when a release must not ship
without its receipts.
- uses: Orphograph/Orphograph/integrations/github-action@master
with:
paths: "dist/*"
Action documentation →
AI agents
MCP server
Give any MCP-capable agent the ability to anchor files and folders it produces. A successful tool response can add a receipt to the agent's output; tool availability does not prove the agent invoked it.
Set up the MCP server →Autonomous agents
Agent audit trail
A dependency-free CLI plus a drop-in skill for OpenClaw-style agents. The agent anchors one-line action records, a daily manifest of its memory files, and any skill file it is instructed to anchor. An action-text receipt commits to the text supplied on stdin; it does not observe the action, prove that it happened, identify the actor, or show the action was authorized or correct.
python3 orpho_agent_anchor.py anchor-memory ~/workspace echo "sent invoice #42" | python3 orpho_agent_anchor.py anchor-textAgent audit-trail documentation →
Datasets
Dataset provenance
Commit a dataset bundle, supporting documents, and acquisition log to one path-bound Merkle root. Online mode sends manifest metadata but not file bytes. The receipt does not prove lawful sourcing, model training, approval, or deployment.
Dataset provenance →Build your own
Everything above sits on one small API: hash locally, POST /api/anchor,
keep the receipt. Free tier is three anchors per day; an
API key or a pack lifts the limit.
Webhooks can tell your systems when a proof pins to Bitcoin —
see webhook docs. Running a platform that RECEIVES receipts? See accepting receipts.