TrustGraph

Browse security & privacySecurity and verification

Security and verification

Build around device signatures, independent pins, verified state, and explicit evidence boundaries.

TrustGraph combines device-signed operations, issuer-signed statements, and locally verified path proofs. Your application chooses what to accept and keeps the keys and state needed to make that choice.

Keep signing authority on the device

Generate a dedicated P-256 key for each device. Browser helpers create a non-extractable private CryptoKey; persist it with IndexedDB structured cloning. Native integrations inject a signing adapter backed by their key store.

Persist a pending bind before opening AuthenSee. It contains an initiation capability. The callback carries a separate completion capability; finalize requires both and the matching local key. Remove the callback fragment after safely retaining it for completion, and exclude both capabilities from analytics, URLs you share, and logs.

Sign the exact operation

Use the SDK's canonical signing helpers. Sign all query members exactly as submitted, before defaults. Each target and each retry needs a fresh signed read and nonce. Writes and reads use distinct signing domains.

Read authorization has a five-minute freshness window. Read replay tracking is process-local; mutation replay handling is separate. Application retry logic must create fresh read authorization instead of replaying captured requests.

Scroll diagram horizontally

The fields bound by a signed readtrust-graph-read-v1operationgraphKeyIdnonce + signedAtExact submitted queryCanonical JSON → P-256 signature → authorization
Sign the exact operation and submitted query with the device key, a fresh nonce and signedAt. Send those same fields in the request.

Hold your own verification pins

Supply the expected Ed25519 attestation JWKS from trusted application configuration. Use the bundled circuit verification key with the optional /zk verifier, which verifies proof bytes locally. Release its WASM resources with destroy() when finished.

Check the meaning of the evidence

A verified positive proves that a policy-satisfying path exists in the committed graph. A negative is a signed server report. A content signature plus membership certificate proves different facts and requires separate current-status and recipient-reachability checks.

Use what a proof tells you when designing labels and authorization decisions.

Does not establishThe proof does not establish that the service published the complete graph. The service sees the graph and queries. A path proof hides intermediate nodes and edges from the verifier; it does not prove graph completeness, latest state, human identity, message signing time, or live presence.

Verify and retain device history

A parsed device directory is not yet verified authority. Verify records, signatures, head hashes and continuity against retained pins. Continue pagination to the advertised head and persist the resulting verified state atomically. A bind can return a quarantined device; use the current verified state when deciding which actions to expose.

See identity and devices API for approval, recovery, revocation and reset behavior.

Keep observability separate

Avoid recording device secrets, bind capabilities, and plaintext signed messages. Show failures with stable API codes where available, and keep rejected verification evidence out of a successful-connection UI. See the API overview for errors and rate limits.

For integration or security questions, contact the team.

Positive answers use the UltraHonk ultrahonk-wot-path-v2 proof scheme; negative answers remain signed reports.

Checked against server 9f47f2d · SDK 12053abLast updated 2026-11-04Edit on GitHub