Introducing April 2026
Your keys work till they don't

The activation-limited API key proxy that gives AI agents exactly the access they need — and nothing more.

Scroll

Every agent gets the keys to the kingdom

When you hand an AI agent an API key, you're giving it unlimited, indefinite access. One bad loop, one hallucinated retry, one prompt injection — and your bill spirals while the agent burns through your quota.

"We woke up to a $47,000 OpenAI invoice. The agent ran overnight."
82%
of developers
ship production API keys with no spend limit
$4.7B
lost annually
to leaked or runaway API credentials
14min
average time
for an exposed key to be exploited
0
providers offer
activation-count limits natively

What if every key had a kill switch?

A proxy that wraps any upstream API key with hard limits — by activations, tokens, or dollars — and stops the moment any limit is hit. No configuration. No dashboards to watch. Just a key that works till it doesn't.

One proxy.
Three limits.
Zero key storage.

Till sits between your AI agent and the upstream provider. You create a scoped key with hard limits. The agent uses it like any normal API key. When the limit is reached, the key is done.

Encrypted at creation

Your real API key is AES-256-GCM encrypted into the scoped token itself

Real-time tracking

Every call updates activation, token, and spend counters instantly

Instant cutoff

Whichever limit hits first kills the key — no grace period, no overrun

1
Create a scoped key — set max activations, tokens, or dollar spend
2
Hand it to your agent — works as a drop-in API key replacement
3
Till proxies every call — decrypts, forwards, and tracks usage
4
Limit reached? — key is automatically exhausted. Done.

Pick your guardrail. Or pick all three.

Whichever limit triggers first exhausts the key. Simple, predictable, impossible to overrun.

Activations

Cap the total number of API calls. Perfect for task-scoped agents that should complete in N steps.

max_activations: 50

Tokens

Limit total input + output tokens. Ideal for controlling context-heavy workloads and long conversations.

max_tokens: 100000

Dollars

Set a hard spend ceiling. Built-in pricing for 80+ models means accurate cost tracking per call.

max_spend: $5.00

Zero upstream keys at rest

Till never stores your real API key. It's AES-256-GCM encrypted and embedded directly inside the scoped token. If someone breaches the database, they get counters — not keys.

AES-256-GCM encryption — military-grade symmetric cipher with authentication

Key-in-token architecture — the scoped key IS the encrypted upstream key

Database stores only metadata — activation counters, limits, and status flags

Revoke instantly — mark a key as revoked and it's dead, no propagation delay

Anatomy of a Scoped Key

till_sk_a1b2c3d4e5f6.U2FsdGVkX1+vuppp
I7aMorHJtFaC2Yp3
mZxQ9sKzRnUwE...
Prefix identifier
Database lookup key
AES-256-GCM encrypted payload
Dot separator

One proxy. Every model.

Twelve providers, 80+ models, one unified proxy layer. Switch providers without changing your integration.

O
OpenAI
A
Anthropic
G
Google
OR
OpenRouter
M
Mistral
G
Groq
T
Together AI
F
Fireworks
P
Perplexity
D
DeepSeek
X
xAI
C
Cohere

Built-in pricing tables for 80+ models — accurate spend tracking out of the box

Start free. Scale when ready.

No credit card required. Upgrade as your agent fleet grows.

Free
$0 forever

Perfect for prototyping and solo agent projects.

  • 3 scoped keys
  • 1,000 activations / month
  • All 12 providers
  • Token + dollar limits
  • Dashboard access
  • Python SDK
Scale
$99 / month

For organizations with large agent deployments.

  • 100 scoped keys
  • 250,000 activations / month
  • All 12 providers
  • Token + dollar limits
  • IP allowlisting
  • Dedicated support
Your keys work till they don't
Get started free