Build production coding agents on a single multi-provider key

The hidden tax of multi-provider LLM stacks

Production coding agents almost always need more than one provider: Claude for long-context reasoning, GPT for some tool-use edge cases, Haiku or Gemini Flash for cheap classification. Wiring all of these up means three SDKs, three rate-limit policies, three billing surfaces, three sets of error handling, three rotation policies for keys. The integration burden grows with every model you add — and so does the cost, because direct-provider pricing leaves no margin for the gateway you'll inevitably build internally.

Anvat exposes both OpenAI and Anthropic wire formats on one key

Call /v1/chat/completions for OpenAI-shape requests; call /v1/messages for Anthropic-shape requests. Same key, same auth header, same base URL. The router picks the right downstream provider based on the model name you specify. Add any frontier model to your stack by changing one string — no new SDK, no new key, no new bill. And every request is 30% cheaper than provider-direct.

Why this beats the obvious alternative

• Single key, both API shapes

  OpenAI SDK works against /v1/chat/completions; Anthropic SDK works against /v1/messages. One env var.

• 30% margin to spend on quality

  Discount lets you afford Opus where you'd otherwise have to downgrade to Sonnet.

• Tool use, vision, streaming all passthrough

  No feature gaps. If the provider supports it, Anvat forwards it.

• Per-tenant cost attribution

  Tag requests with metadata headers, get aggregated spend per customer in the dashboard.

• Crypto + card payment for agent SaaS

  Top up balances with USDC/USDT or Stripe — useful when your agent's end-customers are in regions where cards are friction.

Quickstart

TypeScript — one key, both SDKs

// Anthropic SDK against /v1/messages
import Anthropic from "@anthropic-ai/sdk";
const anthropic = new Anthropic({
  baseURL: "https://api.anvat.app/v1",
  authToken: process.env.ANVAT_API_KEY,
});

// OpenAI SDK against /v1/chat/completions — same key
import OpenAI from "openai";
const openai = new OpenAI({
  baseURL: "https://api.anvat.app/v1",
  apiKey: process.env.ANVAT_API_KEY,
});

// Route per task
const plan = await anthropic.messages.create({
  model: "claude-opus-4-8",
  max_tokens: 4096,
  messages: [{ role: "user", content: planningPrompt }],
});

const codegen = await openai.chat.completions.create({
  model: "gpt-5",
  messages: [{ role: "user", content: codegenPrompt }],
});

FAQ

Can my agent stream responses through Anvat?

Yes. Streaming is fully passthrough — SSE for OpenAI shape, Anthropic's stream events for the /v1/messages path. Cursor, Claude Code, and custom agents all work without buffering.

How do I track cost per end-user in a multi-tenant agent SaaS?

Add custom headers (X-Customer-Id, X-Tenant-Id) on each request. Anvat indexes them and surfaces per-tag spend aggregates in the dashboard — useful for chargeback or usage caps. Headers are forwarded to the provider but logged at the gateway for attribution.

What if I need a model Anvat doesn't host yet?

Email support — we add frontier models within days of provider release. For long-tail open-source models, point to OpenRouter for those specific calls and keep Anvat as your primary gateway. The two key model pattern works fine.

Does Anvat support function calling / tool use?

Yes for both OpenAI tools and Anthropic tool_use blocks. All structured tool-call semantics are forwarded — including parallel tool calls and tool_choice constraints.