Metadata-Version: 2.4
Name: versyn
Version: 0.2.0
Summary: Cryptographic proof for AI actions — sign AI decisions with Ed25519, verify offline with hash binding.
Project-URL: Homepage, https://versyn.dev
Project-URL: Repository, https://github.com/hasoon1231/versyn
Project-URL: Bug Tracker, https://github.com/hasoon1231/versyn/issues
Project-URL: Public Key, https://versyn.dev/.well-known/versyn-pubkey.txt
Author-email: Versyn <hello@versyn.dev>
License: MIT
Keywords: agent,ai,audit,compliance,cryptography,ed25519,verification
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Security :: Cryptography
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Requires-Python: >=3.9
Requires-Dist: httpx>=0.27.0
Requires-Dist: pydantic>=2.0.0
Requires-Dist: pynacl>=1.5.0
Provides-Extra: cli
Requires-Dist: rich>=13.0.0; extra == 'cli'
Requires-Dist: typer>=0.12.0; extra == 'cli'
Provides-Extra: dev
Requires-Dist: mypy; extra == 'dev'
Requires-Dist: pytest-httpx>=0.30.0; extra == 'dev'
Requires-Dist: pytest>=8.0.0; extra == 'dev'
Requires-Dist: ruff; extra == 'dev'
Description-Content-Type: text/markdown

# versyn

**Cryptographic proof for AI decisions.** Sign every AI decision with Ed25519, verify it offline against a pinned public key. 100% coverage, zero per-check cost. Trust the math, not a server.

```bash
pip install versyn
```

## The problem: the "trust tax"

Most AI-monitoring tools audit decisions by sending each one to *another* large model to judge it. That costs money per check, so teams sample — they audit 1% and hope the other 99% was fine. Every un-audited decision is a blind spot a regulator or a lawsuit can walk into.

versyn removes that tax. Each decision gets an Ed25519-signed certificate computed locally. Verification is a signature check on your own machine — no external model call, no per-check fee. You can certify **100%** of decisions instead of a sample, at effectively zero marginal cost.

## Verify a decision offline

```python
from versyn import VersynClient

client = VersynClient(api_key="vk_...")

event = {
    "kind": "agent.decision",
    "payload": {"agent": "underwriter", "action": "decline", "reason": "DTI_too_high"},
}

cert = client.certify(event)
print(client.verify(cert, original_event=event))   # True — local, no network
```

`verify()` runs entirely on your machine. It checks that the signature is valid against the pinned public key, **and** that the certificate hash binds to *your* event. Change one field and verification refuses it.

## Get a free key

```python
import versyn
key = versyn.register("you@company.com")   # 500 certifications/month, free
```

## It survives outages

If the API is unreachable or credits run out, `certify()` queues the event to a durable local file (atomic writes, thread-safe) instead of losing it. On restart the client reloads the queue; `flush_queue()` settles it later. No silent gaps.

## Verify against the published key yourself

The pinned key ships in the package and is also published at:

```
https://versyn.dev/.well-known/versyn-pubkey.txt
```

## What this is — and isn't

A versyn certificate proves a specific decision was signed at a specific time and has not been altered since. It is **tamper-evident, independently verifiable evidence**.

It does **not**, by itself, prove the decision was correct or lawful, and it is **not** regulatory compliance certification. It is the verifiable evidence layer your governance and auditors build on top of.

## License

MIT
