AirspaceObserveAgent Detection and Response

Every AI agent in your company. Who uses it, what it touches, who answers for it.

Airspace is Convoy's Agent Detection and Response. It finds the AI agents working inside your company, from the assistant someone wired into Microsoft 365 or Google Workspace to the server nobody remembers, and puts them all on one screen. Decisions are written as sentences you can answer with one button. The score is five tests, and anyone can recompute it from the register.

Start free · 14 days Make a decision

Governance score · measurement 1

24/100

Exposed

Becomes 71 (Governed) once four decisions are made and the proposed guardrails are switched on.

Accountable27 Decided9 Visible82 Contained0 Guarded0 after the open decisions

Try it live

Four decisions take it from Exposed to Governed.

Eleven agents on three orbits around the company. The closer to the center, the sooner someone has to act. Four are waiting on a decision written in plain words. Make one and watch the score climb. Nothing is enforced until you also switch on a guardrail.

Airspace · the briefLarkspur Group

24/100 · Exposed

Becomes 71 (Governed) once the 4 open decisions are made and the proposed guardrails are on.

Accountable27

Decided9

Visible82

Contained0

Guarded0

Decisions that need you · 4

CIO · Nov 6

An AI assistant is reading the CIO's mail and 50 meeting transcripts.

This week it read 470 emails across 20 mailboxes.

Limit what it readsAllow it

IT lead · Nov 3

Two servers are calling an AI API and nobody owns them.

360 calls in the last 6 hours.

Block until claimedFind the owner

413 people use AI tools · 11 agents act for them · 8 without a named owner

Observe · AirspaceAgent detection & response

Search across assets, identities, rules…⌘K

Convoy platform nominalUTC {{ utc }}Local {{ loc }}Tick T+0000

Active tenantLarkspur Group1 act now· 1 decide todayRisk 27/100 · Guarded

SectorHealthcareRegionNAAssets2,746

Switch tenant⌘O

AI agents at Larkspur Group

Who uses them, what they can reach, who is accountable, and what has been decided. Measured monthly.

Copy summaryPrint brief

BriefRegisterDecisionsGuardrails

{{ score }}/100

Accountable{{ dAcc }} Decided{{ dDec }} Visible82 Contained{{ dCon }} Guarded{{ dGua }} after the open decisions · the rule for each score is printed in the brief

Agents around LarkspurCloser to the center, sooner to act

An AI assistant is reading the CIO's mail and 50 meeting transcripts.

25 people connected it to Gmail over MCP. This week it read 470 emails across 20 mailboxes.

CIO · Nov 6

An AI assistant is reading the CIO's mail and 50 meeting transcripts.CIO · Nov 6

An AI assistant is reading the CIO's mail and 50 meeting transcripts. {{ d1choice }}You, {{ d1when }} · Undo

Two servers are calling an AI API and nobody owns them.

One is an automation server with mail access. The other is unidentified. 360 calls in the last 6 hours.

IT lead · Nov 3

Two servers are calling an AI API and nobody owns them.IT lead · Nov 3

Two servers are calling an AI API and nobody owns them. {{ d2choice }}You, {{ d2when }} · Undo

An agent answers email as a developer, every 10 minutes.

4,348 runs this month under his account. If it is disabled or phished, the agent goes with it.

CIO · Nov 13

An agent answers email as a developer, every 10 minutes.CIO · Nov 13

An agent answers email as a developer, every 10 minutes. {{ d3choice }}You, {{ d3when }} · Undo

5 people connected an AI tool to the production ERP.

A buyer, a shipping rep, a talent manager, an IT PM and the SVP of Operations. No app owner on record.

SVP Ops · Nov 13

5 people connected an AI tool to the production ERP.SVP Ops · Nov 13

5 people connected an AI tool to the production ERP. {{ d4choice }}You, {{ d4when }} · Undo

Now do it on your own tenant {{ stepHint }}

For the board

The six questions the board will ask, answered from your own data.

Straight from the register. The answers update as decisions are made, and the numbers are the same ones in the register.

Who is using AI agents?

413 people across 40+ departments, from IT to the production floor. 68 of them build agents or automations. 16 admin accounts use AI tools.

What can the agents reach?

5 of 11 agents touch sensitive systems: company mail (470 emails read this week across 20 mailboxes), meeting transcripts (50), files, the production ERP and source code.

Is company data leaving?

Yes, to three AI vendors, mostly through sanctioned use: 94 computers calling one, 10 computers plus two unowned servers calling another, 81 computers on the third.

Who is accountable?

3 of 11 agents have a named person or team. 8 have no owner you could call. Two of those are servers calling an AI API on their own.

What have we done about it?

0 of 4 decisions made. 0 of 6 guardrails switched on. Nothing has been blocked. Every decision and guardrail change will be listed here with who made it and when.

Can we see all of it?

Identity logs alone show 7 of the 11 agents. Convoy sees all 11 by adding network and computer telemetry. The remaining gap is laptops without Convoy Agent.

AI code editor

Sign-in logsConvoy

Coding agents, API keys

Sign-in logsConvoy

Cloud coding agent

Sign-in logsConvoy

Direct API use

Sign-in logsConvoy

API keys skip the sign-in, so none of this appears in identity logs. Convoy sees it on the network, and on every computer running Convoy Agent.

How it works

One picture: who uses AI, through what, touching what.

Airspace reads four things you already have: who signed into which AI app, what the Microsoft 365 or Google Workspace audit log says agents did, every AI service call the firewall saw, and what the programs on each machine actually ran. Convoy Agent on the host closes the last gap. Put together, every agent gets a name, a reach, an owner or a blank where one should be, and a decision with a date on it.

The map

People, agents, and what they touch.

Click anything and it moves to the center with everything it touches arranged around it: who is on the left, what it runs on and where data goes on the right. The overview puts agents on three orbits around the company. Closer to the center, sooner someone has to act.

Decisions

Written as sentences, answered with buttons.

"An agent answers email as a developer every 10 minutes." Give it its own identity, or leave it. Each decision comes with a proposed owner and a due date, updates the register and the score the moment it's made, and lands in a log with who and when.

DECISION 5 people connected AI to the ERP OWNER SVP Ops · due Nov 13 [ Review access ] [ Allow it ]

The score

Five tests, no model.

Accountable, Decided, Visible, Contained, Guarded. Each is scored 0 to 100 by a rule printed beside it, then averaged equally. The brief shows what the score becomes once the open decisions are made, so the board sees the gap and the path at the same time.

ACCOUNTABLE named person or team DECIDED a call has been made VISIBLE seen on network or host CONTAINED sensitive agents decided GUARDED Ask first or Stop is on

Guardrails

Rules for agents, off until you say so.

Each guardrail comes from something Airspace found, like a connector reading meeting transcripts or a server with no owner calling AI APIs, and shows how often it would have fired. Pick one of three settings: Watch only logs it, Ask first holds the action until someone approves, Stop blocks it. Nothing is enforced until you switch it on.

Count your agents before the board asks.

Connect Microsoft 365 or Google Workspace, your firewall and your identity provider, push Convoy Agent, and the first measurement is on the screen within a day. 14 days is enough for the first brief and your first decisions.

Start free · 14 days Book a briefing