$0 / month
For individual evaluation and small workloads.
- 10,000 gateway runs / month
- 5 workflows
- 3 agents
- 2 seats
- 2 MCP servers
- Semantic cache eligible for shadow, not serving
// transparent plans
Start with proof, not a contract.
Create a workspace, use an observe-mode key, and see how Punk behaves on your traffic. New verified accounts receive a 30-day Pro trial with no credit card required.
After the trial, the workspace returns to Free unless you choose Pro. Model-provider and third-party service charges are separate from the Punk plan.
$99 / month
Pro — for teams operating production agents.
- 500,000 gateway runs / month
- 100 workflows
- 50 agents
- 10 seats
- 20 MCP servers
- Semantic cache serving available
Custom
For larger-scale, security-reviewed deployments.
- Plan limits tailored to scope
- Direct pilot and rollout planning
- Security and data-boundary review
- Commercial terms agreed directly
- Implementation-specific controls validated during review
Plan limits are code-defined current defaults. Usage and creation limits are enforced when billing is enabled. Self-hosted or billing-disabled development environments may operate differently.
// what counts
A run is work served through the gateway.
Cached and reusable-workflow responses still count toward the monthly run quota because Punk serves and explains that work. A request blocked by the quota does not consume another run.
Provider spend
Your OpenAI, Anthropic, OpenRouter, or other configured provider usage is not included in the Punk subscription price.
Observe mode
Observe-mode traffic counts as gateway runs. Estimated ghost savings remain separate from realized savings.
Trial activation
Pro trial limits activate after email verification. Before verification, effective limits remain on Free.
// common questions
Pricing answers.
Is a credit card required for the trial?
No. Create a workspace and verify your email to activate the 30-day Pro trial.
What happens at the end of the trial?
Your workspace returns to the Free plan unless you choose Pro. The current Free plan includes 10,000 monthly gateway runs.
Does Punk include model tokens?
No. Provider and external-service charges are separate. Punk records available cost information and can identify or serve eligible cheaper routes.
Why does a cached response count as a run?
The runtime still receives, governs, routes, serves, and explains the request. The plan meters gateway work rather than provider calls alone.
Are Enterprise controls guaranteed by this page?
No. Deployment-specific security, retention, redaction, legal, and operational controls must be validated and agreed for the environment in scope.
Use your own traffic as the test.
Start observe-only, inspect the opportunity, and pay for a larger plan only when the runtime is useful to your team.