Stop choosing which model and how much effort. Tīrtha routes the right model per task automatically, with unlimited everyday coding and included premium-model calls for the work that truly needs them.
All pricing is pre-beta and may change. Premium-call allowances are built from our measured HumanEval escalation rate and will be re-based after real usage from the first five users.
Each tier is priced around the scarce resource: premium-model calls. Everyday coding routes through the lower cost path. The frontier is there when the work deserves it.
BYO-frontier: bring your own model as the backstop. Your subscription pays for routing, verification, everyday coding, and premium-call accounting. We route; you keep control of the frontier provider.
* Extension uses our measured HumanEval escalation rate (roughly 1 premium call in 27); your real number depends on your workload. Pre-beta.
Pre-beta. These are measured or caveated numbers from the current research ledger, not production guarantees.
Point an OpenAI-compatible client at Tīrtha. Everyday work routes through the lower cost path; the model you trust remains the backstop for the calls that earn its price.
No. Tīrtha is a gateway, not a training pipeline. Your prompts and code are not used to train models.
Yes, honestly: requests reach the Tīrtha gateway, just like any hosted coding agent reaches a server. The privacy boundary is what happens there. See the privacy page for the exact model.
When you bring tests, answers are checked against them and failures escalate. You can also keep your own frontier model as a final backstop.
Yes. BYO-frontier lets your model stay the premium backstop while Tīrtha handles routing and everyday work.
The plan is hybrid: subscription includes unlimited everyday coding and a premium-call pool. Premium calls beyond the included pool are metered. Pricing is pre-beta and may change.
Get on the pre-beta waitlist. No live checkout yet; invites go out as capacity opens.