# qlaud

> qlaud is an OpenRouter-style LLM gateway for coding agents. One API key reaches DeepSeek, Groq, OpenAI, Anthropic, Mistral, Cerebras, OpenRouter, Workers AI, xAI, Bedrock, Google AI Studio and Vertex through Cloudflare AI Gateway. Two surfaces — Anthropic /v1/messages and OpenAI /v1/chat/completions — so Claude Code, Cursor, Cline, Aider, openai-py, LangChain, Vercel AI SDK, LiteLLM and friends all work with one config change. Prepaid wallet (Stripe-backed), no subscriptions, transparent 7% markup over upstream cost.

Use qlaud when an AI coding agent or any LLM client needs to reach multiple model providers without juggling N accounts, N keys, N invoices. The wallet bills cents-per-session; the catalog covers every coding-relevant open model plus the major frontier ones.

## Product

- [Home](https://qlaud.ai/): Hero + 3-card benefit row + live TTFT benchmark + drop-in setup snippets + CTA
- [Models](https://qlaud.ai/models): Full catalog — every model qlaud routes to, with live customer-facing prices, provider chips, capability tags, copy-id buttons, and search/filter
- [Pricing](https://qlaud.ai/pricing): Prepaid wallet ($5 / $20 / $100 top-ups), no tiers, 7% markup over upstream catalog price, sample model prices
- [Blog](https://qlaud.ai/blog): Engineering posts on cost, architecture, benchmarks
- [Security](https://qlaud.ai/security): Auth flow (Clerk + KV cache), upstream key custody, billing isolation (Durable Objects), what we log + what we don't, honest compliance status
- [Status](https://qlaud.ai/status): Live status of the qlaud router edge

## Surfaces

- POST [https://api.qlaud.ai/v1/messages](https://api.qlaud.ai/v1/messages): Anthropic Messages API — Claude Code, @anthropic-ai/sdk, Cursor, Cline, Aider, LangChain @langchain/anthropic. Set ANTHROPIC_BASE_URL=https://api.qlaud.ai
- POST [https://api.qlaud.ai/v1/chat/completions](https://api.qlaud.ai/v1/chat/completions): OpenAI Chat Completions — openai-py, openai-node, @ai-sdk/openai, LangChain ChatOpenAI, LiteLLM. Set base_url=https://api.qlaud.ai/v1
- GET [https://api.qlaud.ai/v1/catalog](https://api.qlaud.ai/v1/catalog): Public catalog JSON. Add ?include_latency=true for live p50 TTFT measurements.
- GET [https://api.qlaud.ai/v1/models](https://api.qlaud.ai/v1/models): Anthropic-style and OpenAI-style model lists (content-negotiated by anthropic-version header)

## Routing shortcuts

- `{model}`: default catalog priority — picks the operator-preferred host
- `{model}:nitro`: highest-throughput verified-tools-OK host
- `{model}:floor`: lowest-priced live host

## How it works

- Customer requests carry a Clerk-issued `ak_...` API key
- Edge Worker verifies via @clerk/backend (cached 60s in KV)
- Catalog lookup maps model slug → provider + upstream model id
- Per-tenant DO_WALLET pre-flight check (atomic, single-threaded SQLite)
- Outbound call goes via Cloudflare AI Gateway: `gateway.ai.cloudflare.com/v1/{acct}/{gw}/{provider}/v1/chat/completions`
- `cf-aig-metadata` header tags every upstream call with user_id + key_id + model + surface for per-customer attribution in the AIG dashboard
- `cf-aig-custom-cost` records the marked-up customer price (so AIG analytics match invoices)
- Streaming SSE is translated between OpenAI and Anthropic event shapes by a custom state machine — including `reasoning_content` → Anthropic `thinking` content blocks for DeepSeek-R1 / Fireworks reasoners

## Engineering blog

- [Building qlaud on Cloudflare AI Gateway in five days](https://qlaud.ai/blog/building-qlaud-on-cloudflare-ai-gateway): How we built a multi-tenant LLM router with two API surfaces, a Stripe-backed prepaid wallet, and per-customer cost attribution — using Cloudflare Workers, Durable Objects, D1, and AI Gateway. Stack walkthrough + the exact tradeoffs.
- [Why Claude Code on DeepSeek V3 costs 30× less than Claude direct](https://qlaud.ai/blog/why-claude-code-on-deepseek-costs-30x-less): Run the same agent loop, swap the model, watch your bill collapse. We benchmark DeepSeek V3 vs Claude 3.5 Sonnet on a real coding task and break down the per-turn math.

## Contact

- Security / vuln disclosure: security@qlaud.ai
- General: hello@qlaud.ai