Till Bounded access
for AI agents Request access

The control layer for agentic work

Your keys work
till they don’t.

Give every AI agent a key with a finish line—before it starts working.

See why 9 panels · 4 minutes
Access passTILL / 001
Ready for work
Requests
500hard stop
Spend
$25budget
Expires
6 PMtoday

Not your provider credential.
Not open-ended permission.

The uncomfortable default

An agent gets the whole key.
The task needed a sliver.

01

Too much permission

A provider key can keep spending and calling long after the job should be done.

02

Too little context

The credential cannot tell a prototype, contractor, production agent, or one-off task apart.

03

One blunt response

When something goes wrong, teams rotate the shared secret and disrupt everything attached to it.

A smaller permission surface

Don’t hand the agent your master key.
Hand it permission to finish one job.
BeforeProvider key

Broad · shared · open-ended

With TillScoped access

Bounded · disposable · revocable

The shortest path to control

Three moves.
One hard stop.

  1. 01

    Connect

    Save your existing AI provider credentials as encrypted connections.

  2. 02

    Bound

    Issue a Till key with the exact limits the work deserves.

  3. 03

    Run

    Point the workload at Till. Requests route to your configured provider.

  4. STOP

    Limit reached

    Later requests are rejected before another provider call is dispatched.

The finish line is yours

Set what ends
the permission.

Every key has a required request ceiling. Add the other boundaries the work calls for.

Required500

Requests

A fixed number of provider calls.

Optional100K

Tokens

Optional$25

Spend

Optional6 PM

Expiry

Optional1 IP

Network

Any time×

Revoke

Designed for a smaller blast radius

Your provider key
stays behind Till.

  • Encrypted connectionsAES-256-GCM at rest with bound context.
  • Disposable agent keysThe workload never needs the underlying provider credential.
  • Pre-dispatch enforcementCapacity is reserved before a provider request leaves Till.
  • Immediate local revocationStop one workload without rotating every integration.
Read the security model

Bring the providers you already use

One control layer.
Twelve providers.

Automatic keys can route among the connections configured for your account. Till tracks activations and provider-reported usage through one surface.

  • 01OpenAI
  • 02Anthropic
  • 03Google
  • 04OpenRouter
  • 05Mistral
  • 06Groq
  • 07Together AI
  • 08Fireworks AI
  • 09Perplexity
  • 10DeepSeek
  • 11xAI
  • 12Cohere

250+ priced provider/model identifiers for estimated-spend enforcement. Provider usage remains billed by your provider.

Straightforward beta pricing

Start bounded.
Scale deliberately.

Provider usage is billed separately through your own provider accounts. Till is currently onboarding approved beta customers.

Free$0/ month

Prove the workflow.

  • 3 scoped keys
  • 1,000 activations / month
  • All 12 providers
Request access
Most useful
Pro$59/ month

Run active agent workflows.

  • 25 scoped keys
  • 25,000 activations / month
  • All limits and providers
Choose Pro
Scale$249/ month

Standardize across more work.

  • 100 scoped keys
  • 250,000 activations / month
  • All limits and providers
Choose Scale

Founding customers who joined before July 18, 2026 retain documented price protection. See full pricing terms.

Till

The job has an end. Its permission should too.

Give your next agent
a finish line.

Request beta access Read the docs