One API for
every model.
Point your existing OpenAI client at one endpoint and reach OpenAI, Anthropic, and more. Switch models by changing a single string - with automatic routing, failover, and one prepaid bill.
Full privacy - developed & hosted in Switzerland
- ~40ms
- Added p50 latency
- 99.9%
- Routed uptime
- 1 bill
- Across providers
// point your existing OpenAI client at Bold
const client = new OpenAI({
baseURL: "https://boldrouter.com/v1",
apiKey: process.env.BOLD_KEY
});
await client.chat.completions.create({
model: "openai/gpt-5.6-sol",
messages,
stream: true
});One integration. Every major provider.
Everything between your code and the model
A thin, fast data plane on the hot path. A full control plane for keys, billing, and analytics - never in the way of a request.
Unified OpenAI-compatible API
Full /v1/chat/completions with streaming SSE, tools, and multimodal pass-through. Zero SDK changes.
Smart routing & failover
Route by price, latency, or uptime. When a provider is down or rate-limited, we fail over automatically.
One prepaid balance
Buy credits once, spend them at provider list price. One small fee on top-ups, one bill across every provider.
Token-accurate metering
We count usage from the stream itself, handle interrupted requests, and reconcile against provider numbers.
Scoped API keys
Issue, scope, and revoke hashed keys. Per-key spend limits and live usage, all from the dashboard.
Usage analytics
Per-key and per-model spend, tokens, and latency - with price comparison across the catalog.
The journey of a single request
Format translation is the bulk of the work - and we keep it lean so the proxy adds barely-there latency.
- 01
Normalize
Your OpenAI-format request is parsed and validated once at the edge.
- 02
Route
We pick the best provider for the model by price, latency, and live health.
- 03
Translate
The request is rewritten to the provider's native format - system prompts, tools, multimodal blocks.
- 04
Stream & meter
Deltas stream back as OpenAI chunks while tokens are counted for billing.
Compare, then switch with one string
Transparent pricing across the catalog. Every model is callable through the same endpoint.
| Model | Context | Input / 1M | Output / 1M |
|---|---|---|---|
GPT-5.5 openai/gpt-5.5 | 400K tok | $5.00 | $30.00 |
Claude Opus 4.8 anthropic/claude-opus-4.8 | 1M tok | $5.00 | $25.00 |
Gemini 3.5 Flash google/gemini-3.5-flash | 1M tok | $1.50 | $9.00 |
Grok 4.3 x-ai/grok-4.3 | 256K tok | $1.25 | $2.50 |
DeepSeek V4 Pro deepseek/deepseek-v4-pro | 1M tok | $0.44 | $0.87 |
GLM-4.6 zai/glm-4.6 | 200K tok | $0.43 | $1.74 |
Provider list price - your credits buy tokens at cost, with no per-token markup.
Pay only for what you route
No seats, no minimums. Prepaid credits, metered per token.
Pay as you go
For everyone, from prototype to scale.
Tokens at provider list price - no per-token markup · BYOK billed at 5% of usage
- Every model, one endpoint
- Automatic routing & failover
- Streaming, tools, multimodal
- Unlimited API keys
- Usage analytics & price comparison
Enterprise
For high-volume and regulated teams.
- Volume-based platform-fee discounts
- BYOK - bring your own provider keys
- SSO, audit logs, SLAs
- Dedicated routing regions
- Priority support
Frequently asked questions
What boldrouter is, which models it routes to, and how billing works.
What is boldrouter?
boldrouter is a unified, OpenAI-compatible API gateway to every major large language model provider. You send requests to a single endpoint with one API key, and boldrouter routes them to OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, Perplexity, Z.ai, MiniMax, NVIDIA, or Qwen - with automatic failover and a single prepaid bill across all of them.
Which LLM providers and models does boldrouter support?
boldrouter supports eleven providers through one API: OpenAI (the GPT-5 family), Anthropic (Claude Opus, Sonnet, and Haiku), Google (Gemini), xAI (Grok), DeepSeek, Mistral, Perplexity, Z.ai (the GLM models), MiniMax (the M-series), NVIDIA (Nemotron), and Qwen (Alibaba). You target a model by its catalog id - for example openai/gpt-5.5, anthropic/claude-opus-4.8, qwen/qwen3.7-max, or minimax/minimax-m3 - or use an auto-router id such as bold/auto to let boldrouter pick the best model for each request.
How is boldrouter different from OpenRouter?
Like OpenRouter, boldrouter is an OpenAI-compatible gateway to many providers with model fallback and provider routing. boldrouter adds opt-in prompt anonymization (PII is redacted before any provider sees your request), built-in document input (PDF, Word, and Excel are extracted to text), token-accurate metering that bills even interrupted streams, and a single prepaid balance charged at provider list price with no per-token markup.
Is boldrouter compatible with the OpenAI API and SDKs?
Yes. boldrouter implements the OpenAI /v1/chat/completions, /v1/embeddings, and /v1/models endpoints exactly, so any OpenAI SDK works unchanged - you only change the base URL to boldrouter and use your boldrouter API key. Streaming, tool and function calling, and vision are all supported.
How does boldrouter pricing work?
boldrouter is pay-as-you-go with no seats and no minimums. You buy prepaid credits and spend them at each provider's list price, with no per-token markup; a small fee (5%) applies when you top up credits. With bring-your-own-key (BYOK), the provider bills you directly for tokens and boldrouter charges only a thin platform fee on that usage.
How does model routing and failover work?
You pick a model or an auto-router id, and boldrouter selects the best provider by price, latency, or uptime. If a provider is down, rate-limited, or returning errors, boldrouter automatically fails over to the next route - across both providers and models - using a health-aware circuit breaker, so a single provider outage does not break your application.
Can I use my own provider API keys with boldrouter (BYOK)?
Yes. boldrouter supports bring-your-own-key for every provider. Your keys are encrypted at rest, the provider bills you directly for tokens, and boldrouter charges only a thin platform fee on that usage.
Can boldrouter read files like PDF, Word, and Excel?
Yes. Attach a file to a chat request - PDF, Word (.doc/.docx), Excel (.xls/.xlsx), RTF, or plain text and CSV - and boldrouter extracts its text before routing, so document input works with every model, even ones without native file support.
Does boldrouter anonymize or redact sensitive data?
Yes. boldrouter offers opt-in prompt anonymization: it detects PII such as emails, phone numbers, credit-card numbers, and names, replaces each with an opaque placeholder before any provider sees the request, and can restore the original values in the response. The mapping exists only for the duration of the request and is never logged or stored.
Does boldrouter store my prompts?
boldrouter logs only routing metadata for each request - model, provider, token counts, and latency - never your message content. With prompt anonymization enabled, PII is additionally redacted before requests ever leave the gateway.
How much latency does boldrouter add?
boldrouter adds roughly 40 milliseconds at the median (p50), most of which is translating your request into each provider's native format, and it targets 99.9% routed uptime by failing over across providers automatically.
How do I get started with boldrouter?
Create an account, generate an API key in the dashboard, point your OpenAI client's base URL at boldrouter, and start sending requests. Target a specific model by its catalog id, or use bold/auto to route automatically across providers.
Ship with every model today
Create a key, change one base URL, and you're routing. No new SDKs, no per-provider billing.
