markupLlama 4↖ go on, drag them around
Switching AI models should take one line of code. Not one quarter of procurement, five vendor reviews and an API key living in a Slack thread.
Four jobs, one platform. Each one is a vendor you no longer have to evaluate. Keep scrolling, they stack.
Change models by changing one word.
All 60+ models sit behind one standards-compatible API. When a better model ships next Tuesday, your engineers edit a string. Nobody files a new procurement ticket.
const res = await tarqa.chat({
- model: 'gpt-4.1',+ model: 'claude-opus-4-6',
messages
});A spending limit that actually says no.
Every request is checked against request and token limits before it reaches a provider. Cross either line and Tarqa returns a 403. Alerts are nice. Blocking is better.
Answers from your documents. Not from vibes.
Index PDFs, sites and repos into a workspace knowledge base. Pass one parameter and responses come back grounded and cited. No separate vector database to buy, secure or babysit.
The GPU you already paid for, finally earning its keep.
One CLI command exposes a self-hosted model to the whole workspace through an encrypted reverse tunnel. Internal testing runs on your hardware at $0 cloud inference.
We know you'll ask. Here's the short version. The long version is a call away, and we'll bring the documents.
Request the security pack →Pick a model, hit send. The request body barely moves.
{
"model": "claude-opus-4-6",
"messages": [{ "role": "user",
"content": "What can Tarqa do?" }],
"rag_context": "workspace-docs"
}You pay providers for tokens, directly. You pay us for the control plane. That's the whole model.
A gateway and control plane that sits between your applications and every AI provider. You register, route, govern and observe every model through one endpoint instead of stitching together SDKs, keys and dashboards for each cloud.