“We’ve recently started using Vigil to optimize API costs for the AI agents our team uses in development, and the results have been impressive. We’ve managed to reduce our costs by roughly 40%, which makes a real difference for a growing company like ours. It gives us more flexibility to experiment, build new things, and scale without worrying as much about API expenses.”
Your agents, up to 90% cheaper to run*
Vigil sits between your app and your AI provider and stops you paying for tokens you don’t need.
Used by the vigilant.
VIGILANT, N. WHOEVER MONITORS AND OPTIMISES THEIR AI.“Our biggest operating costs are tokens. Vigil has slashed that cost in half, allowing us to give waaay more value to users for less. This is a serious competitive advantage, and I’m glad I took the time to implement it :)”
“While we were testing our AI coach, Vigil showed us where we were burning tokens we didn’t have to. Now that we’re heading into production, it keeps the cost per call down. It’s the tool I wish we’d had from day one.”
“Most of my money goes to tokens. I use AI all day to build admins and the systems that run my business, across three projects that mostly live on it. Vigil makes my AI load only the tools it needs, not everything, every time. It’s a small thing that has helped me a lot. It was worth the time it took to set up and see how much you are actually saving; for me, it was a game changer.”
“Vigil saved us roughly 43% on our company’s API traffic, and the monitoring helped us track down a couple of hidden errors in our agents. My developers had spent ages looking for a token optimisation tool, and Vigil is the first that does both monitoring and optimisation, at an affordable rate. The team was also very responsive: when we hit errors in our agents’ backend, they helped us solve everything quickly. Definitely worth trying.”
In customers’ own words. Your savings depend on your traffic.
One URL. Every call. Every cent.
Your SDK. Your provider. Now in view.
Open the quickstart ↗Three optimisations, each cutting what’s left.
Explore the dashboard ↗Prompts and responses pass through in memory. They are never written to disk.
What Vigil sees and stores ↗
baseURL → api.vigil.tools/YOUR_ID/anthropic
X-Vigil-Agent: research-scoutYour provider keys stay where they are.Example. Each cut applies to what’s left. Shadow mode measures your own numbers.
Your prompts pass through. Only the numbers stay.
We keep
One row per call, such as- Cost and tokens
- Speed
- Model and status
- Your agent’s name
- Tool names, the one exception
We never keep
Never written down- Your prompts
- The model’s replies
- Your provider API keys
- Tool inputs and results
- Error text
- Stored in the EU
- No other account can read it
- Delete your call data any time
- Flags risks without copying the text
See every call. Tune every agent.
Monitor every call. Optimise what it costs.
Rendered from the actual dashboard components.
Sample workspace, both sides. Preview controls are inactive.
Free monitors every agent. Paid switches optimisation on.
Start free, see what caching would save you, and let your own numbers make the case.
Free
No card$0
Optimisations On until $30 saved, once.
- Monitoring on every agent
- 10,000 logged calls a month
- 7 days of history
- 1 seat
Pro
Recommended$99/mo excl. VAT
Optimises up to $500 saved per month.
A limit, not a promise. What you save depends on your traffic.
- Monitoring on every agent
- 2,000,000 logged calls a month
- 90 days of history
- Unlimited seats
- Threats and PII pages
Enterprise
Custom
Optimises up to a limit agreed in your contract.
- Monitoring on every agent
- Unlimited logged calls
- Custom retention
- Unlimited seats
- Threats and PII pages
Questions
What counts as saved?
Only measured savings on calls where Vigil applied an optimisation: today that is prompt caching. Shadow estimates never count: they are what an optimisation would have saved, not what it did. Routing and history trimming savings aren’t counted toward your limit yet.
What happens at the saved limit?
We tell you at 80% and at 100%; at the limit optimisations stay On for a 30-day grace period, then run in Shadow until your billing period ends. Never Off; requests never fail. Vigil keeps forwarding, logging and pricing every call. Free’s $30 is once ever: after it, the same 30-day grace, then Shadow until you upgrade.
What happens past the monthly call cap?
Every call still goes through and is still counted toward the month, but Vigil stops storing the call itself until your next billing month starts (on Free, the next calendar month), so it does not show in your call history. Nothing is charged for those calls, and optimisations keep running as before.
Can I change plan at any time?
Yes, from Settings → Billing: upgrade from Free, switch between Plus, Pro and Supreme, or cancel. Your limit follows your plan, counted over the same billing month. Cancelling keeps your plan to the end of the period you have paid for.
What happens if I cancel?
Your plan stays active until the end of the period you have paid for, and you are not charged again. Then the account drops to Free: monitoring continues on every agent, and optimisations follow Free’s rule, On until Vigil has saved you $30 in total, then a 30-day grace, then Shadow. Cancelling does not delete your account or your data.
From setup.
To first signal.
See how to connect your app to Vigil.
SETUP WALKTHROUGH · 23 SECONDSGo deeper.
What does Vigil store from my prompts and replies?+
Vigil reads requests and responses in transit to calculate costs and detect issues. It stores call metadata, not full prompt or response text. Claude-format tool names and prompt fingerprints are among the derived data it keeps. See our security page for the complete breakdown.
What happens to my calls if Vigil is down?+
Vigil is in the request path. If the proxy itself is unavailable, requests can fail. You can point your SDK back to your provider directly. Internal monitoring failures are designed to fail open where possible; they are different from a full proxy outage.
How much latency does Vigil add?+
Vigil adds a network hop and request processing. We do not publish a universal latency figure: it depends on your provider, region and request. Test it on your own traffic before expanding your rollout.
What counts as saved?+
Only measured savings on calls where Vigil applied an optimisation count toward the saved limit. Today that is prompt caching. Shadow estimates show what you could save, and do not count as money saved.
Can I switch caching off per agent?+
Yes. Choose Off, Shadow or On for each agent. Shadow measures potential savings without applying caching changes. You decide when to switch it on.
Keep building.
We’ll keep watch.
One URL. Your numbers. Your decision.