What is Proofpane?

Proofpane is an AI governance platform — the evidence plane for governed AI work. Every AI action across a team's tools is gated by policy at runtime, recorded in a tamper-evident hash-chained audit log, and exportable as an Ed25519-signed Evidence Pack that an auditor verifies offline, with no Proofpane account and no need to trust the vendor.

Who it's for

The people who answer for the company's AI — what it spends (CFO), what it leaks (CISO / internal audit), what it decides (CEO / board). Teams keep their existing tools; the tools and the workflow don’t change for the people who just use them.

What it does

Gates before execution: allow / deny / human-approval on AI tool calls; DLP redacts secrets before a model sees them; risky actions pause for a menu-bar approval.
Covers the tools teams already use: Claude Code, Cursor, Codex, Claude Desktop and MCP clients via each vendor's official hooks and MCP (never TLS interception); n8n / UiPath / Zapier via an egress gateway and execution-audit webhook; any OpenAI-compatible API agent.
Meters and caps spend: the organisation's monthly budget refuses its workflow steps, agent turns and skill runs once spent, an agent's own cap refuses that agent's turns and the skills it runs, and a user's budget set to block refuses that user's agent turns once it is spent; each refusal comes before the model is called and is audited. Calls proxied through the egress gateway are metered but not capped; a department's budget set to block refuses an agent chat's turns once it is spent, and on every other route whether it refuses a call is not yet measured; per-model totals to check against the vendor invoice.
Proves what happened: hash-chained, append-only audit with a DB-level immutability trigger and Ed25519 chain-head anchoring; 335 controls pre-mapped across NIST AI RMF, ISO/IEC 42001, EU AI Act, GDPR and SOC 2.
Headless by design: end users never open a Proofpane screen — their own AI software talks to Proofpane machine-to-machine (MCP, an agent-to-agent A2A API, CLI, egress gateway); dashboards and the menu-bar approval tray exist for oversight.
Improves under the same governance: a skill's model default changes when a named person accepts a comparison recorded on your own cases, or when an auto policy applies one classed improved; the agent's memory method changes only on a significant result — tested on your data, reversible.

What it is not

Not a policy-document / GRC checklist tool — the evidence is generated by runtime enforcement, not attested after the fact.
Not a wrapper IDE or replacement agent — governance rides official extension surfaces.
Not TLS interception — no root CA, no new attack surface; per-client coverage depth is published honestly instead.
Not a certification — Proofpane produces operational evidence; it does not replace legal advice, certification bodies, or regulator judgment. See the Trust Center.

How it is distributed, and whether it is signed

macOS — signed and Apple-notarised. Every macOS artifact distributed today carries the PROOFPANE LIMITED Developer ID (team B94QM75QNG): the daemon (both architectures) and the public Tray universal .dmg, which is also stapled.
Windows — not signed, because Authenticode is a separate certificate that has not been bought. Linux — SHA-256 digests rather than a platform-signing claim.
Three separate rows: the Windows answer does not carry over to macOS. Per-artifact record with hashes and verification commands: releases.json · install & verify · Trust Center.

See it

Live demo — populated org, no signup · walkthrough recordings · install · how it differs from GRC and log tools

Last updated 2026-08-09 · Proofpane, Auckland, New Zealand · [email protected]