More choice, better performance, and the freedom to use the right AI model for every task.
One person, own keys or ours, no invoice you did not trigger.
Start freeThree to fifty people who need one rule set and one bill.
Start a teamOrganisations that need the relay inside their own perimeter.
Talk to usYou pay for the AI you use, with access to leading models from Claude, GPT, Gemini and more. Your balance is used across your AI requests, while many platform features cost nothing extra.
Use leading models at their current provider rates, with no hidden markup.
Routing, memory, caching, security checks and team features are built into the platform.
Open models, cached requests, blocked requests, memory reads and more don't consume your balance.
Choose the models that fit the task, switch whenever you need, and keep building without a subscription.
$30 a seat a month, no other charge. That funds your team's shared pool. Need more than it covers in a given month? Add funds anytime.
If an answer here changes your mind about us, it was going to change your mind eventually.
A flat monthly fee punishes light months and subsidises heavy ones. 12% of paid usage means a quiet week costs almost nothing, and it keeps our incentive pointed the same way as yours — we make less when the router picks a cheaper model, which is exactly what it is built to do.
It would earn us more, and it is the obvious thing to worry about. Every routing decision is written to the log with the reason, the tier it started on and whether it escalated, so the pattern is auditable. You can also pin models and cap tiers, which removes the discretion entirely.
Each seat brings a share of paid model usage and the shares pool, so a heavy week for one person draws on the quiet weeks of others. You see the pool fill in the dashboard, get a warning well before the edge, and hit a hard stop rather than an overage bill. Every seat funds $30 into that pool.
For most teams, yes. You pay the vendor directly at your rate and keep any committed spend discount. The trade is setup and rate limits: you manage a key per vendor and inherit each vendor's throughput.
Paid models stop. Open models keep working, the relay keeps running, and the log keeps recording — nothing is lost and nothing overdraws. On Team the hard stop is the same mechanism, applied to the pool.
Developer to Team is adding seats to the account you already have; keys, memory and history stay. Team to Enterprise is a deployment change rather than a data migration. Downgrades take effect at the end of the billing period.
No. Requests are relayed and logged for your own auditing, and the retention window is yours to set. On self hosted deployments nothing leaves your network at all.
Neither, exactly — the open models are free permanently, not for fourteen days. You can run the relay against them indefinitely without funding a balance, and add money the day you want a frontier model.
Developer needs a key and nothing else. Team can be added later without moving anything — the same account grows seats. Enterprise starts with a conversation about where it runs.
The operational layer between your team and every AI model it touches.