Entropa_

Glossary

Every term used on this site and in Entropa's product, explained in plain English. No background in AI, blockchains, or cryptography assumed — if a word here confuses you, that's a gap in this page, not in you.

The app
Entropa Gate
The app you install on your own Linux machine. It limits what an AI can read, checks what leaves, asks a person to approve big actions, and keeps a signed record of every decision.
Egress Guard
The part of the app that checks everything leaving toward a model or tool, and blocks or removes secrets, personal data and your do-not-send terms. It is still being improved; its test results, including failures, are on the results page.
Access tier
A level of access. Public, staff, agents and builders each get only what their role needs.
Capability token
A short-lived, signed pass that allows one narrow thing. It can only be narrowed, never widened, and it is never shown to the AI.
Checkpoint
A signed summary of the whole ledger at a moment in time. Anyone can check that later records did not change earlier ones.
Witness
An optional separate service that also signs the ledger's checkpoints, so nobody can quietly rewrite history. Extra proof, not required.
Offline verifier
A small program that checks the signatures and proofs on your own machine, without trusting us or the network.
Sealed test set
A set of test cases written by someone other than the builder and kept secret until one scored run. It is then spent and never reused, so results cannot be tuned to it.
Apache-2.0
The planned open-source licence for the code: anyone may use, change and share it, with an explicit patent grant and no warranty. Not yet in force.
The basics
AI agent
A piece of software that can make decisions on its own — reading information, deciding what to do, and taking an action — without a human clicking a button for every single step.
Sequencer
The single piece of software that receives submitted actions, puts them in order, and appends them to the log. Deliberately simple — one writer, one order, nothing to coordinate between independent parties.
Attestation (also: record)
One recorded, signed fact — "this hash existed, at this time, and here's cryptographic proof of it." This is the actual product: $0.01 buys one attestation.
Receipt
The proof document you get back for a specific attestation — your record's content hash, its position in the chain (leaf index), and the block hash linking it to every record before it. You (or an auditor) can re-derive the hash chain independently to confirm nothing was altered.
The log itself
Merkle log
An append-only list of records, summarised by one short value (the root) that changes if any record changes. Each signed checkpoint covers the root, so tampering with an earlier record is detectable, and anyone can check that a record is included without downloading the rest.
Record (also: leaf)
One entry in the log — the payload's content hash, its position (leaf index), the timestamp it was sequenced at, and the block hash linking it into the chain.
Hash (also: fingerprint, digest)
A short string of letters and numbers computed from a piece of data, unique to that exact data. Change even one character of the original data, and the hash comes out completely different — so a matching hash proves nothing was altered.
Using the product
Scryon
Entropa's read-only explorer (scryon.entropa.space): browse the signed ledger and check signatures and proofs yourself, with no login.
Compliance terms you'll see
HIPAA
A US law governing how healthcare information must be protected and handled.
HITRUST
A widely-recognized security certification, common in healthcare, that proves an organization meets a rigorous, independently-audited set of data-protection controls.
SOC 2
An independent audit report confirming a company has real, working controls around security, availability, and confidentiality of the systems it runs.
EU AI Act
European Union law regulating AI systems, with extra requirements for "high-risk" AI — including, under Article 12, automatic and tamper-resistant logging of what the AI system did.
NIST AI RMF
The US National Institute of Standards and Technology's voluntary framework for managing AI risk — it calls for documented, verifiable records of what an AI system actually did.
DORA
The EU Digital Operational Resilience Act, which sets ICT risk and logging requirements for financial entities and their providers.
PCI DSS
The Payment Card Industry Data Security Standard. It requires, among other things, protected audit logs and an inventory of cryptography in use.
NIST SP 800-53
A US government catalogue of security and privacy controls, including controls on protecting audit information.