Agent governance

The system of record
for AI agents.

Coding agents and copilots do real work on real machines now. Proxara observes them at the device, seals sensitive data inline, and writes every action to a signed record.

Request a demo Explore the docs

Coverage where agents already run

Claude CodeCodex CLIGemini CLICursorClaude DesktopWindsurfCopilotAider

An inventory that assembles itself.

There is no registration step. An agent runs and the record begins: the product, the device, the person, the session. Inventory is observed, not self-reported.

Agents · this quarter

claude_code1,284today

gemini_cli412today

cursor388yesterday

codex_cli1first observedjust now

tool_use · notion.append_blocksealed on device

"page_id": "9f2c…",
"content": "Met [CLIENT_NAME] about [ACCOUNT_NO]"

By product · one signed decision

claude_codeAllowed

unknown_sdkBlocked

Sealed before it leaves the device.

Sensitive values swap for stable tokens inside the request itself. The model sees the token. The agent keeps the real value locally, and the work never pauses.

Governance stays one decision deep: allow or block, per product, signed.

A record with nothing to leak.

The ledger stores what ran, never what it carried: the tool, the argument names, the byte counts, the data classes. Signed on the device, chained, and verifiable offline.

Prompt"tighten the q2 billing export"

Tool callnotion.append_blockkeys: page_id, content · 2.1 KBclient_pii

Resultok · 0.9 KB

Signedsha256 8c41…22abverifies offline

Ask Rox. Skip the dashboard.

Rox answers from the signed record in Slack, in Teams, or in the assistant a team already uses. When a security review asks how agents are governed, the answer is already written.

Which agents ran this quarter, and what did they touch?