Why our pricing
looks like this.
Pricing is a hypothesis to validate with design partners, not a final answer. Here is the reasoning behind every decision — and what we are deliberately deferring.
We sell memory, not verification
The CLI verifies. The platform remembers, audits, and proves. A free command-line tool can check a build once; it cannot give a team a shared history of how faithful their software has been to its specifications over time. We align price with the thing customers cannot rebuild themselves.
Open core
Free forever: the CLI, the GitHub Action, bring-your-own-key, unlimited local evaluations, MIT licensed. Paid: the hosted platform, dashboard, evaluation history, audit trails, team collaboration, and governance. This is how GitHub built a business on Git and Datadog on open-source monitoring. We draw the line deliberately: the commodity (the CLI) is MIT, and the defensible layer (the dashboard, audit, and governance) is proprietary — so no cloud provider can resell the valuable part.
Bring your own key
Customers connect their own model provider key and pay providers directly. We are not reselling tokens, so a check carries almost no marginal cost — which protects margins and means we never ration checks. It also matches the privacy promise exactly: code stays on the customer's machine and is sent only to the provider they choose.
The positioning ladder
The category we claim grows as we earn the right to claim it. We lead with the wedge and grow into the larger category, rather than overreaching on day one.
- Stage 1 now we are hereSpec Fidelity
Does the software match the intent? A clear, novel, ownable wedge that avoids the crowded code-quality category.
- Stage 2 nextThe Audit Layer for AI-generated software
Provenance, traceability, accountability, governance. A materially larger category.
- Stage 3 eventuallyThe System of Record for AI software delivery
Institutional memory, historical truth, compliance, organizational trust. Credible only once the records exist.
What we watch first
Pre-product-market-fit, the pricing table is the least important thing on the page. MRR and ARPU are lagging outputs. These four leading indicators decide whether the model works:
- Time to first hosted Spec Fidelity report (activation).
- Repositories per account (adoption).
- Seats per account (expansion).
- Weekly active accounts (retention).
Operating rules
No fabricated metrics, testimonials, logos, or customer counts appear on the pricing page or anywhere else. We do not imply customers or numbers we do not have. The first-25 cap on the Design Partner program is real and will be honored.