Pricing
Start free. Scale as you grow.
Every tier includes dashboards, approval workflows, and the contextual bandit. You're never optimizing blind, and nothing goes live without your sign-off.
Starter
For indie developers and early startups evaluating the platform.
Get started- 10K decisions/month
- 1 experiment at a time
- Up to 3 arms per experiment
- Standard LLM for ideation
- Basic personalization
- Hourly model refits
- 30-day data retention
- Community support
Pro
RecommendedFor growing startups with real traffic and one product surface to optimize.
Get started- 250K decisions/month (then $0.50/1K)
- 5 simultaneous experiments
- Up to 8 arms per experiment
- Frontier LLM with extended thinking
- Full interaction modeling
- All ideation heuristics
- Composite rewards and guardrail metrics
- Variable-based personalization
- Unlimited data retention
- Conclusions with reasoning
- Email support
Growth
For scaling companies optimizing multiple surfaces that need to prove ROI.
Get started- 2M decisions/month (then $0.30/1K)
- 25 simultaneous experiments
- Unlimited arms (system-managed)
- Ensemble model decision-making
- Deep feature engineering
- Advanced convergence (faster learning)
- Holdout and incrementality measurement
- Cross-experiment learning
- Surrogate models for long-term reward prediction
- Webhooks and warehouse exports
- Priority compute and support
- Team seats and roles
Enterprise
For high-traffic companies with compliance requirements and proprietary data.
Talk to us- Unlimited experiments and volume
- Third-party data enrichment
- Pre-loaded CRM and customer data
- Custom synthetic data generation
- Custom reward distributions
- Identity resolution and CDP integration
- SLA-backed latency
- Dedicated environments
- SSO, audit logs, RBAC
- Custom contracts, security review, DPA
- Dedicated solutions engineer
Comparison
Compare every plan.
| Feature | Starter | Pro | Growth | Enterprise |
|---|---|---|---|---|
| Volume | ||||
| Decisions/month | 10K | 250K | 2M | Unlimited |
| Overage rate | — | $0.50/1K | $0.30/1K | Negotiated |
| Simultaneous experiments | 1 | 5 | 25 | Unlimited |
| Arms per experiment | 3 | 8 | Unlimited | Unlimited |
| Data retention | 30 days | Unlimited | Unlimited | Unlimited |
| Model refits | Hourly | More frequent | Priority compute | Priority compute |
| Personalization | ||||
| Contextual bandit | ||||
| Basic personalization | ||||
| Full interaction modeling | — | |||
| Deep feature engineering | — | — | ||
| Variable-based personalization | — | |||
| Identity resolution / CDP | — | — | — | |
| Intelligence | ||||
| Standard LLM ideation | — | — | — | |
| Frontier LLM with extended thinking | — | |||
| Core ideation heuristics | All | All | All | |
| Ensemble decision-making | — | — | ||
| Surrogate models | — | — | ||
| Cross-experiment learning | — | — | ||
| Advanced convergence | — | — | ||
| Custom synthetic data | — | — | — | |
| Optimization | ||||
| Composite rewards | — | |||
| Guardrail metrics | — | |||
| Custom optimization instructions | — | |||
| Holdout / incrementality | — | — | ||
| Custom reward distributions | — | — | — | |
| Trust & visibility | ||||
| Dashboards and visualization | ||||
| Approval workflows | ||||
| Conclusions with reasoning | — | |||
| Third-party data enrichment | — | — | — | |
| Pre-loaded CRM data | — | — | — | |
| Infrastructure | ||||
| Webhooks / warehouse exports | — | — | ||
| Dedicated environments | — | — | — | |
| SLA-backed latency | — | — | — | |
| SSO, audit logs, RBAC | — | — | — | |
| Support | ||||
| Community | — | — | — | |
| — | — | — | ||
| Priority | — | — | — | |
| Dedicated solutions engineer | — | — | — | |
| Team seats and roles | — | — | ||
FAQ
Questions, answered.
What counts as a decision?
Every call to /decide is one decision. If you show an optimized upgrade prompt to a user, that's one decision. Tracking events (/events, /identify) don't count.
What happens when I hit my limit?
On Starter, decisions after 10K return your control content (no optimization, no disruption). On Pro and Growth, overage pricing kicks in automatically. We send alerts at 50%, 75%, and 100% so there are no surprises.
Can I review what the AI generates before it goes live?
Yes, on every tier. Approval workflows let you review and approve AI-generated arms before they reach your users. This is on by default and you can turn it off when you're comfortable.
What does "simultaneous experiments" mean?
An experiment is one element being optimized (your upgrade prompt, your homepage headline, your onboarding email). Simultaneous experiments is how many you can run at the same time.
How does the contextual bandit differ from A/B testing?
An A/B test splits traffic evenly and waits for statistical significance. A contextual bandit shifts traffic toward what's working as it learns, and personalizes per user based on context. It's faster, wastes less traffic on losing variants, and gives each user the version most likely to work for them.
Do I need Growth to run multiple experiments?
Pro supports up to 5 simultaneous experiments. Growth raises that to 25 and adds cross-experiment learning, where insights from one experiment inform others.