Heimdall
The LLM gateway our own products run on — keyed, metered, provider-agnostic.
The problem
Once more than one application calls a model, cost stops being attributable and the vendor SDK becomes a dependency you cannot get out of. Both problems are gateway-shaped.
What we built
Every product we run calls models through Heimdall rather than through a vendor SDK. It issues per-tenant keys, meters usage, and means a model swap is a config change instead of a deployment.
Per-tenant keys
Each application gets its own key and its own usage ledger, so cost is attributable to the thing that spent it.
Provider-agnostic
Swap the model behind an application without touching the application.
Metering built in
Token and request accounting per tenant, which is the number you actually need before you can price anything.
Runs in production
Live at heimdall.caffinix.com and carrying the traffic for every product on this page.
Pricing
Two ways to buy it.
Heimdall can run on our infrastructure under your brand, or on yours. The difference is who gets the call when it breaks.
per branded instance · $1,200 setup
- Deployed under your brand, on your domain
- We host, monitor, patch and update it
- A failure at 2am is ours to fix, not yours to diagnose
- Monthly — cancel when it stops being worth it
+ $600/yr updates and support
- Runs on your own infrastructure, inside your own network
- One production instance, plus as many staging and local copies as you need
- Yearly renewal covers updates, security patches and support
- Source code is not included in either tier
Build something
with the same stack.
Bring us the workflow, agent, or product backend you need shipped. We will map the fastest credible build.