Build on Tīrtha.
One base URL, your existing client. Start with the quickstart, then the reference. Our policies are below, in full.
Tīrtha is in public beta. Keys and limits may change. Billing follows the prices listed.
Quickstart →
Get a key, then make your first request with curl, PowerShell, Python or JavaScript.
API reference →
Endpoints, request fields like mode and no_cache, the tirtha label on every response, and errors.
Setting up other tools →
Set up OpenCode, Cline, Roo Code, Kilo Code, Continue, Aider, Cursor, LangChain or LiteLLM, each from its own docs (Cursor from its community forum).
Server info →
Limits, spend caps, cache policy, latency, and the known gaps.
The Tīrtha router →
Send tirtha/auto: the levels, what the check does and does not do, billing, and the label on every response.
Capabilities →
What is measured, generated from the live configuration and our claims ledger, including what we have not measured.
Tīrtha Bench →
Every model we have tested on the same 800 tool-calling tasks and trap prompts, with the same scorer, misses included.
Security →
What we never do, what is standing guard right now, and what you control: revocable keys and enforced spend caps.
Trust →
Where everything lives, and why there is no SOC 2 yet and what we do instead.
Where your code goes →
What reaches our gateway, what the shared cache stores (answers with their prompt), and how to opt out per request.
Terms of Service →
How the service works, your content and ownership, AI output accuracy, acceptable use, fees, and warranties.
Privacy Policy →
What we collect, how it is used, how long it is kept, and your rights over your data.
Changelog →
What changed on Tīrtha, and when.