Pricing and FAQ
Pricing and FAQ
Pricing
What does it cost?
During a pilot we agree the price with you. The first month is free; then it becomes a paid plan or ends.
The guarantee
What exactly can’t prompt injection change?
The action, the recipient and the records read. It can change the words in a draft and which allowed option the model picks.
Doesn’t the model still read the attack?
Yes, and it may obey. The design assumes it does, which is why the guarantee doesn’t depend on the model.
Who writes the code?
A model writes each version’s definition once, from your routine’s instructions: Cloudflare Workers AI on Cresec’s model allowance, or Anthropic under your workspace’s own key.
Is this proven?
It’s the code-then-execute pattern from published prompt-injection research, including Google DeepMind’s CaMeL. We publish our threat model.
What doesn’t it cover?
Open-ended agents, chains where one agent calls another, and stolen user sessions.
For security teams
Why would we say yes?
Cresec can help make an approval discussion concrete: which person, tool version, app and action is allowed? IT still needs to assess the host, connector support and operating controls. Review enrolled tools and the activity Cresec actually mediates. The view covers enrolled tools and their gateway-mediated calls. Document the enrolled workflows and routed actions during setup.
What happens if Cresec is unavailable?
Cresec’s gateway requires a fresh control-plane decision for each governed call. When the control plane cannot be reached, every governed call, reads included, is refused with an explicit, retryable error; nothing falls back to a stored credential or a cached decision. Deployment review covers error reporting, expiry and revocation of cached display data, and recovery. Resolve a timed-out write’s outcome before retrying it.
What do signed records prove?
Check the integrity of signed records for calls routed through Cresec. Verification covers the records present under the stated key and export assumptions. Assess activity coverage separately through reconciliation and failure testing. Independently retained receipts anchor verification to their last checkpoint.
Where does the model run?
Turning a routine into a tool runs on Cloudflare Workers AI on Cresec’s model allowance, or on Anthropic under your workspace’s own key. Clef on Cloudflare Workers AI handles a tool’s declared free-text choices. Workers AI requests may be processed outside the United States.
How do we stop using Cresec?
You retain your workflow; plan and measure the work required to leave. You can restore direct vendor access under your own credential management. For your workflow, we measure the code and configuration changes, credential migration and elapsed time with you.
For builders
Do I need to code?
You do not need to code to describe the job or explore the sample. A real integration may require someone who can change the workflow, configure identity and review its permissions.
Will teammates see my emails or customers?
Gmail and CRM calls through Cresec use the person’s own connected account. Review the workflow’s caches, shared outputs and vendor permissions too; credentials alone do not prevent every data leak.
Can the tool send on its own?
Sending or changing business records requires a person-specific grant in Cresec’s default policy. Creating a draft in Gmail is itself an upstream write and needs the right OAuth scopes. Tool/version approval, each person’s consent and send/write grants are separate decisions. A changed action set can need review.