Trigger 01
A surprise invoice
A provider bill lands that's twice last month's, and nobody can say which agent did it.
Governance Snapshot: every agent and API key, owned and costed, in one week See the offer
Agent Control Plane
AI spread through your company faster than anyone could track it. We find every agent and API key, give each one an owner and a budget, and keep the register current, without touching a single prompt.
Trigger 01
A provider bill lands that's twice last month's, and nobody can say which agent did it.
Trigger 02
An API key is still spending money, and the person who created it left in the summer.
Trigger 03
"List every AI system that touches our data." Due Friday.
Trigger 04
ISO or SOC auditors arrive, and AI now sits inside the scope.
Trigger 05
A provider sets a deprecation date. Which of your agents run on that model?
Trigger 06
"What AI are we running, and what does it cost?" There's no good answer yet.
Teams wired OpenAI and Anthropic keys into n8n flows, internal bots and product features. Each one made sense on its own. Put together, nobody can say what's running, who owns it, or what it costs. (If everything you run is on flat monthly subscriptions, this isn't for you.) Provider consoles show totals per key. They don't show spend per agent or per owner.
All agents 23
€0 / month
| Agent | Team | Provider | Owner | € / mo |
|---|---|---|---|---|
| support-triage-bot | Support | Anthropic | Ana K. | 412 |
| invoice-extractor | Finance | OpenAI | Unowned | 188 |
| contract-reviewer | Legal | Anthropic | Marko P. | 96 |
| lead-enricher | Sales | OpenAI | Tina R. | 641 |
| search-feature (prod) | Product | Bedrock | Luka B. | 1,204 |
| weekly-report-agent | Ops | Gemini | Eva S. | 37 |
Illustrative data
What you get
“I would love to have a product and have those hours done somewhere else.”
Engineering lead, software agency, after building their own tracker
Controls
The Snapshot sets up the same six controls a company with a full IT team would have built. You get them without asking a developer to build and maintain a tracker.
Every agent, automation and API key across your providers, found from billing and admin data and named in plain language. Includes the ones nobody remembers creating.
A named person on every agent. When someone leaves, their keys don't become orphans that keep spending.
Cost split by agent, team and owner, not one total per provider. You can see which agent made the bill go up.
A budget on each agent and an alert before it's blown. Quiet when everything is fine, so nobody has to watch a dashboard.
A light, written rule for adding new agents: who approves, what data it may touch, what it may spend. Light enough that teams will follow it.
A dated export of the register for the customer questionnaire, the auditor or the board. You answer in an afternoon, not a fortnight.
Where agents hide
Each agent traced from the API key it spends on to the provider and model it calls, and to the person accountable for it.
Automation agent
n8n · finance
Controls
4 / 6 in place
Inventory
Registered
Owner assigned
Spend
Attributed per agent
Budget €200 / mo
Rules
Data scope approved
Model retirement checked
How it works
Four steps: scoping, connect, snapshot, keep current. You always know what's next and who owns it. Your team's part is a few short calls.
Step schematic
ACP Snapshot · Step 1
Involved
Owner + ACP
= one agent-week = needs attention
A 45-minute call with whoever gets asked "what AI are we running?" We list the providers, the teams that build with AI, and the date that's driving this.
You give us read-only billing and admin access to each provider. We never sit in your traffic and we never see a prompt or a response.
We build the register, chase down every unowned key with your team, attribute spend per agent, and write the approval and budget rule with you.
You run the rules from here. New agents and keys come to you for approval before they spend, budgets hold per agent, and owners get updated when people leave. You get an alert before a budget breaks or a model retires.
Built on restraint
No proxy, ever. We connect to the provider APIs that already know what you spend, so nothing sits between your agents and the model and there's nothing to install.
Read the detailsOpenAI
Anthropic
Azure
Bedrock
Pricing
Small enough to approve this week. Every price is fixed before we start.
Compare offers
All prices excl. VAT
€2,500 one-off
Your real workload tested across every model that could do the job. You get today's cost against each alternative, the quality difference on your own cases, and a migration order with the risks flagged.
Ask about the AuditStart here
€1,900 one-off
The complete register: every agent and API key, an owner on each, spend per agent, plus the budget and approval rule for new ones. Ready to hand to an auditor, a customer or the board.
Book a Snapshotfrom €90 / month
After the Snapshot, you govern your agents yourself. Set the rules, approve new agents and API keys before they spend, and keep budgets per agent. The register stays true without anyone maintaining it.
Talk to usBuilt with
Agencies, AI startups and SaaS companies told us how their agents and API keys really spread, and what it took to get them back under control.
FAQ
No. We read billing, usage and admin data through each provider's own read-only APIs. Nothing passes through us, and we never see a prompt, a response or a customer record. You can revoke access any time.
The person who gets asked "what AI are we running and what does it cost?", usually a COO, CFO, head of ops or an IT lead with no developers to spare. It fits companies of roughly 50 to 500 people where agents and API usage spread through teams without IT being involved from the start.
No. Observability tools help engineers debug and tune one application. We answer a different question for the business: what's running, who owns it, what it costs, and what the rule is. Plenty of teams use both.
Flat-priced seats are out of scope. We cover autonomous agents and API-metered usage, where the cost and the risk grow without anyone noticing.
They could, and some teams do. The hard part is keeping it current once they're busy or have moved on. The Snapshot gives you the result in one week. Keep Current means nobody has to maintain it.
In the EU (Frankfurt). We sign a DPA before we connect anything, and the register is yours to export at any time.
AgentControlPlane
Every agent, the key it spends on, the model it calls and the person who owns it, on one page. Start with a 25-minute call and we'll tell you straight whether a Snapshot fits.
or write to hello@agentcontrolplane.com