PROVINGX
Zero TrustFull Proof

Every AI call asks permission first.

Provingx decides whether a model call is allowed before it runs, then signs the decision into a receipt anyone can verify.

Start free Open a live dashboard
No card required Keys out of app code Signed, verifiable receipts
Sits in front ofOpenAIAnthropicGoogle GeminiGroqMistralTogetherAzure OpenAIAny MCP server
How it works

Three steps, then it is running.

POST /v1/chat/completions18 checks
OpenAI
 
 
 
 
 
 
 
 
 
 
 
Anthropic
 
 
 
 
 
 
 
 
 
 
 
01

Change one line

Point your existing client at Provingx. No rewrite, no wrapper SDK, and your provider keys stop living in application code.

Team headers
02

Hand your team one snippet

A shareable header bundle carries the team, the agent and its passport. A teammate fills in only their own name.

Public verification
03Same on both surfaces

Every decision comes back signed

Allowed or denied, each call leaves an Ed25519 receipt that anyone can check against your public key.

Run both and they join up: Intent binding matches a tool call to the model decision that asked for it.

The control room

Everything else it does, doing it.

Fleet Control

Stop everything, now

One button freezes every agent across the org, on the next call, with no redeploy. It works on every plan, including a lapsed trial.

Agent kill switch

Or kill one agent

Halt a single agent, one team, or one accountable user. Everything else keeps running.

Live and sandbox

A sandbox that cannot reach production

A separate partition with its own keys and its own data, one click away.

Enforced Secrets

Secrets it can never print

Register a secret and only its fingerprint leaves your browser — we never hold the plaintext. Any call carrying a match is redacted in flight, in both directions.

Public transparency log

Even we cannot backdate it

Seal the audit log's Merkle root and stamp it to Bitcoin through OpenTimestamps. Confirmation takes hours, so a fresh anchor honestly reads pending — and what it leaves behind is a standard .ots file, checkable with the public client rather than with us.

Compliance evidence

Framework readiness assessed against real traffic, exported as a signed attestation.

Dependency graph

Which agent calls what, built from the traffic rather than a diagram someone drew.

Waste ledger

What you paid for calls that broke your own rules, were discarded, or ran twice.

Chain of custody

Every multi-agent handoff signed, with each delegation provably narrower than its parent.

Everything the platform does
Verify us

Do not take our word for any of it.

Every receipt is signed with your org’s own Ed25519 key. Run the check yourself — our script, your machine, your org’s public key, and no request to us in either direction. Change one number in the file and it says so.

01Get the verifier
$ curl -OJ \ https://api.provingx.com/api/verify/cli
02Get any receipt
$ curl -s \ https://api.provingx.com/api/verify/RUN_ID \ -o receipt.json
03Check it, offline
$ python3 verify_receipt.py receipt.json# exit 1 if one byte moved
Your terminal — verify_receipt.py
How offline verification works

Put a permission layer in front of your AI today.

Free to start, and the stop button works on every plan.

Start free Book a walkthrough