Model access on one API.
One endpoint for many model providers, with failover, per-tenant wallets and a ledger, so every Dirmacs product and every client agent draws tokens from one metered supply.
One supply of tokens, metered per tenant.
Drop-in endpoints
Chat completions, models and embeddings, OpenAI-compatible. Existing code moves over with a change of base URL.
Failover across three providers
Requests fail over across three upstream providers. Rate limits and usage headers come back on every response.
Wallets, ledger, guardrails
Every tenant has its own wallet and ledger, with PII guardrails and caching on the way through.
One endpoint in, metered before it moves.
Your app calls one endpoint. Haimdall meters the call against the tenant's wallet and ledger, routes it to a provider, and fails over if one is down.
- YOUR APP
- HAIMDALL
- PROVIDERS
- WALLET · LEDGER
Our own products first, then developers who need local rails.
Every Dirmacs product
Haimdall is the token supply under everything we run. Each client's model spend lands in its own ledger, so it is attributed automatically.
Developers and teams who need local rails
Pay in rupees by UPI or RuPay, top up from ten rupees, and get a GST invoice on every top-up.
How a top-up is priced.
Provider price plus a small margin, in local currency per million tokens. No subscription. A tax invoice on every top-up.
The token supply for everything we run.
Haimdall is the token supply under every Eruka and ARES deployment we run. One supply, one ledger, and every tenant's spend attributed on its own line, wherever the call originates.
See the platform