Vigil

One URL — monitor and optimise API traffic.

Start freeSee how it works

No card required

Your stack, the moment it ships

vigil.wtf/dashboard

Illustrative

HOW IT WORKS

See everything. Fix anything.

Optimise
1 error

Token optimisation

Vigil caches the repeated parts of your prompts so you pay less — here's what you could save, what you're saving now, and what's already been saved.

DayWeekMonth

Could save

$196.25/ mo

if every Off and Shadow agent were switched On

Saving now

$247.74/ mo

what optimisation is capturing now, at the observed rate

Saved · last 30 days

$247.74

over the last 30 days, on the agents optimisation is switched on for right now

Fleet cache hit · 30d

20%

Modes

On 2 · Shadow 1 · Off 3

Warming up

0

support-triageON60%$172.70 /mo
pr-reviewerON13%$75.04 /mo
research-scoutSHADOW0%$196.25 /mo
+ 3 more agents

Illustrative

Everything you need

See inside your entire AI stack.

Every call. Every agent. Every failure — caught before your users notice.

Monitor

Every call
logged.

Token counts, costs, latency, model, agent — all in one searchable table. Filter by agent, model, date, or status instantly.

Calls
1h6h24h7d30d
TIMEAGENTMODELTOKENSCOSTLATENCY
14:23:07support-botclaude-sonnet-42,847$0.01371.2s
14:23:08onboardinggpt-4o1,203$0.00442.4s
14:23:09lead-qualifierclaude-3-haiku847$0.00110.8s
14:23:10email-respondergemini-1.5-pro3,421$0.00621.9s
14:23:11sales-closergpt-4o1,892$0.00693.1s
14:23:12support-botclaude-sonnet-49,841$0.04732.8s
14:23:13invoice-parsergemini-1.5-flash4,102$0.00041.5s
14:23:14pa-agentmistral-large1,556$0.00412.1s
Detect

Failures caught
before users notice.

Cost spikes, prompt injections, PII leaks, infinite loops — Vigil detects them automatically and tells you exactly what went wrong.

3 active errors
Detected issues requiring your attention.
AllCost spikesSecurityPerformance
COST SPIKEsupport-bot2m ago
Spent 3.5× the expected amount. Stuck in a clarification loop — ran 31 turns instead of the usual 6.
Fix →Mark resolved
PROMPT INJECTIONlead-qualifier5m ago
System prompt keywords detected in user message. Rule #7 matched. Potentially adversarial input.
Fix →Mark resolved
PII LEAKinvoice-parser12m ago
Email address pattern detected in model output. Data exposure risk — customer PII in plain text.
Fix →Mark resolved
Fix

Plain English
fixes.

When something breaks, Vigil explains it in plain English, shows which agent caused it, and gives you the exact code fix.

APPLY FIX
cost_spike · support-bot
Detected 2 min ago · $0.0473/call (baseline $0.0137)
The Support Bot spent 3.5× the expected amount on a single conversation. The agent ran 31 turns instead of the usual 6, stuck in a clarification loop. Adding a max_tokens cap will prevent this from happening again.
Code FixUnderstand WhyTest It
  await client.messages.create({
−   max_tokens: 4096,
+   max_tokens: 1024, // caps the clarification loop
  })
Apply fix →Mark resolved
Connect

Two lines.
Full visibility.

Point your existing SDK to api.vigil.wtf/YOUR_ID/anthropic — your dashboard shows the URL with your real id filled in — add your Vigil key as a header, and every AI call is instantly captured. No wrappers, no new SDKs: a base URL and one header.

anthropic.ts
import Anthropic from "@anthropic-ai/sdk"
 
const client = new Anthropic({
− baseURL: "https://api.anthropic.com"
+ baseURL: "https://api.vigil.wtf/YOUR_ID/anthropic",// route
+ defaultHeaders: { "X-Vigil-Key": process.env.VIGIL_API_KEY },// authenticate
});
Works with Anthropic, OpenAI, Gemini, Mistral, Groq and 12 more providers. YOUR_ID is your account id — the dashboard shows this URL filled in, one click to copy.
Scale

17 providers.
One proxy.

Anthropic, OpenAI, Gemini, Groq, Mistral, Together AI — every provider flows through the same Vigil layer automatically.

AnthropicAWS BedrockGoogle VertexOpenAIGoogle GeminiMistralxAI GrokFireworks AIBasetenCloudflare Workers AIGroqTogether AIDeepSeekMoonshotCerebrasZ.aiOpenRouter

Every provider you add routes through the same Vigil URL automatically. No extra config.

Pricing

Monitoring is free. Forever.

Every plan watches every agent: cost, latency, errors, and a measured figure for what caching would save you. You only pay to switch optimisation on — so start free and let your own numbers make the case.

Get started for free →Explore plans

No credit card · Free forever · Upgrade only to switch optimisation on

Questions about what we store, what we can optimise, or what it costs? Read the FAQ

1 URL changes to set up
100% Of AI calls captured
17 Providers supported
60 Seconds to first insight

Your AI stack is running right now.
Do you know what it's doing?

Free forever — every agent monitored, 100k calls a month. Takes 60 seconds.

Get started free →

No credit card · 60 second setup · 17 providers supported

EXAMPLE ALERTS
✓support-bot — 847 calls logged$2.40