Pricing
Design partnership is where the validation record is load-bearing. Self-serve SaaS exists so you can run the loop on your own data first.
Design partner · enterprise
- → Loop wired against your data, one team, one model class to start
- → All six honest-eval validators on every spec
- → MCGrad subgroup multicalibration on your subgroup definitions
- → Distribution-shift decomposition + OOD detection
- → Audit-ready HTML model card per run
- → Managed deployment; bring-your-own-key for LLM
- → Core workflow placement; multiple teams and model classes
- → Validation record integrates with your eval infrastructure
- → Self-hosted deployment, dedicated GPU pool
- → SAML + SCIM, SOC2 evidence packet, DPA
- → Custom validators specific to your domain
- → Custom promotion thresholds, custom subgroup definitions
- → Strategy-domain preview included (agent-code, RL)
- → Dedicated CSM
Design partners during the early phase get below-band pricing in exchange for co-developing on shared roadmap items. The validation record we produce is yours when the engagement ends.
Self-serve · try the loop yourself
Tabular classification + regression on uploaded CSV / parquet. Honest-eval (all six validators), MCGrad calibration, model card, and the predict endpoint are included on every plan — the differentiator is how much compute we'll let you spend.
- → 5 runs / month
- → 2 datasets · 500 predict calls / month
- → Hyperparameter sweep strategist
- → All 6 validators (shuffled-label, randomized-feature, secondary-holdout, perm-FWER, dist-shift, multi-cal)
- → MCGrad calibration + model card + predict endpoint
- × LLM agent strategist (needs token budget)
- × LLM-authored feature pipeline code (enterprise only)
- → 100 runs / month
- → 20 datasets · 10,000 predict calls / month
- → 1M LLM tokens (Anthropic / OpenAI pool)
- → LLM agent strategist + pre-execute critic
- → Composable feature transforms (split, regex, hash, datetime, text)
- → Everything in Free
- → Email support
- → 1,000 runs / month
- → 200 datasets · 100,000 predict calls / month
- → 10M LLM tokens (or bring your own key)
- → Up to 5 users · SSO (Google OAuth)
- → Slack support
- → Everything in Starter
Frequently asked
What does a design partnership cost?
$30K–$200K/yr for smaller teams running one model class; $200K–$1M+/yr for workflow integration across multiple teams. Below-band pricing during the early phase in exchange for co-developing on shared roadmap items. Talk to us →
What is the validation record?
Every spec, every rejection, every promotion, every calibration update — SHA-256 audit-chained, contract-frozen, queryable. The deliverable that compounds across the engagement; the thing that costs more to rebuild than to license. Co-developed on your data; yours when the engagement ends.
Can I self-host?
Yes, on the enterprise plan. The engine is closed-source; we license it with a support contract. SAML / SCIM, SOC2 evidence, DPA, dedicated CSM. Contact sales →
What's a “run” on the self-serve plans?
One closed propose-run-validate-promote loop, regardless of how many iterations or specs it produces. A 5-iteration HP sweep with 30 specs is one run.
What happens at the run cap?
New runs return HTTP 402 with an upgrade link. In-flight runs still complete and findings remain queryable. No surprise overages.
Do I need to bring my own LLM key?
No. Starter and above include a token allowance against our LLM provider pool. Bring-your-own-key is supported on Team and Enterprise if you want billing on your own provider account.