# Tarqa AI

> Tarqa AI (legal name: Tarqa AI Solutions) is the unified gateway and control plane for AI. Route 60+ models — Claude Opus 4.6, GPT-4.1, o3, Gemini 2.5 Pro, Llama 4 Maverick, DeepSeek-R1, Mistral Large 2, Qwen3 and more — plus your own local GPUs through one OpenAI-compatible API endpoint. BYOK with $0 token markup. Built-in RAG, dual-gate budget governance, 5-role RBAC, SAML 2.0 SSO, and exportable audit trails. Zero vendor lock-in.

## Core pages

- [Homepage](https://tarqaai.com/): What Tarqa AI is, how the gateway works, GPU tunnels, RAG, governance features
- [Pricing](https://tarqaai.com/pricing): Free ($0), Starter ($15/mo), Pro ($49/mo), Team ($149/mo), Enterprise (custom)
- [Documentation](https://tarqaai.com/docs): API reference, authentication, streaming, SDKs, RAG guide, GPU tunnel setup
- [Blog](https://tarqaai.com/blogs): Guides on AI integration, cost optimization, model routing, and enterprise AI
- [Status](https://tarqaai.com/status): Real-time API uptime and service health
- [Legal](https://tarqaai.com/legal): Privacy policy and terms of service

## Key facts

- **One API key for 60+ models** — Claude, GPT, Gemini, Llama, DeepSeek, Mistral, Qwen, Nova, and more; swap models with a one-line change to the model string
- **BYOK, $0 markup** — bring your own provider keys; token usage billed directly by providers; Tarqa charges only for the control plane
- **GPU WebSocket tunnels** — one CLI command exposes any self-hosted or open-weight model on a local GPU as a first-class API endpoint for the whole team; $0 cloud inference cost
- **Built-in RAG** — vector search over your docs, sites, and repos with no separate vector database; add `"rag": "workspace"` to any request
- **Dual-gate budget governance** — hard limits on both request count and token count per key; requests return 403 on breach, making runaway AI bills physically impossible
- **5-role RBAC** — Owners, Admins, Developers, Members, Viewers; per-seat token budgets; workspace isolation
- **SAML 2.0 SSO** — integrates with any identity provider
- **Exportable audit trails** — every token logged; encrypted in transit and at rest; configurable data-retention controls
- **Context-aware memory** — persistent conversation state that survives sessions with token budgeting
- **OpenAI-compatible endpoint** — drops into existing OpenAI SDKs by changing the base URL and model string; no code rewrite required

## Company

- **Founded**: 2025
- **Founder & CEO**: Sudhanshu Tiwari
- **Headquarters**: India
- **Legal name**: Tarqa AI Solutions
- **Brand name**: TarqaAI
- **Website**: https://tarqaai.com
- **GitHub**: https://github.com/Tarqa-Ai
- **LinkedIn**: https://www.linkedin.com/company/tarqaai
- **X / Twitter**: https://x.com/tarqaAI
- **Instagram**: https://www.instagram.com/official.tarqa/

## Pricing summary

| Plan       | Price         | Requests/mo | Tokens/mo | Seats |
|------------|---------------|-------------|-----------|-------|
| Free       | $0            | 100         | 200K      | 1     |
| Starter    | $15/mo        | 1,000       | 1.5M      | 1     |
| Pro        | $49/mo        | 8,000       | 12M       | 1     |
| Team       | $149/mo       | 40,000      | 60M       | 5     |
| Enterprise | Custom        | Custom      | Custom    | Unlimited |

## Comparable products / alternatives

TarqaAI is an alternative to: OpenRouter, Portkey, LiteLLM, Helicone, BerriAI, and direct provider SDKs. Key differentiators: GPU WebSocket tunnels for local models, built-in RAG (no separate vector DB), dual-gate hard spend caps, and $0 token markup with BYOK.