Private, reliable LLM routing
for production AI apps.
Route the open-weight leaders — Qwen, GLM, DeepSeek, Gemma, Kimi, MiniMax — plus every frontier model, through one OpenAI-compatible API. TrustedRouter keeps prompt traffic on an attested gateway, avoids prompt/output logs, and gives teams a verifiable trust path instead of another black-box router.
- 🔒 No prompt logs
- ▤ Metadata-only control plane
- ◈ Failover logic
Production AI now needs an inference control layer.
Most AI products start by calling one model provider directly. That works until you need fallback, model choice, cost control, regional routing, privacy controls, provider-specific policies, customer trust, or proof that sensitive prompts are not being logged by another intermediary.
One gateway. Three jobs.
Route
Send requests to many models through one OpenAI-compatible API. Use explicit models or trustedrouter/auto for provider fallback.
Protect
Keep prompt traffic on an attested gateway. The control plane stores metadata for billing and operations, not prompt or output bodies.
Prove
Publish the source commit, image reference, image digest, and attestation path so technical customers can verify what is running.
Move over with one base_url change.
- ✓ Use
trustedrouter/autofor fallback - ✓ Use
trustedrouter/euwith the EU regional API for Europe-focused routing - ✓ Use
provider.data_collection = "deny"for zero-retention provider preference - ✓ Use BYOK where needed
- ✓ Keep existing OpenAI-shaped calls working
const client = new OpenAI({
apiKey: process.env.TRUSTEDROUTER_API_KEY,
baseURL: "https://api.trustedrouter.com/v1"
})
const response = await client.chat.completions.create({
model: "trustedrouter/auto",
messages: [{ role: "user", content: "Hello" }],
provider: { data_collection: "deny" }
})Your prompt path should be inspectable.
TrustedRouter separates production inference from the normal dashboard/control-plane surface. Prompt and output bodies should not pass through the control plane. Metadata is used for billing, routing, latency, status, and cost tracking.
Fallback when providers fail.
Production AI should not break because one provider returns 429s, 5xx errors, degraded latency, or temporary capacity issues. TrustedRouter routes across healthy providers and exposes route health so teams can build more resilient AI products.
View statusPrivacy is not a setting buried in a dashboard.
For sensitive AI products, the question is not only which model you use. It is who can see the prompts, what gets logged, which providers retain data, and whether the router itself can be inspected.
Built for teams where prompts matter.
Legal AI
Pain: Cannot risk exposing client privilege to intermediaries.
Solution: Route sensitive client work with a verifiable prompt path to zero-retention providers.
Read security details →Finance & PE
Pain: Evaluating private data with models creates non-disclosure risks.
Solution: Evaluate companies, documents, and private data without another opaque intermediary.
Read security details →Production AI apps
Pain: Provider downtime breaks core product reliability.
Solution: Keep production inference reliable across providers using fallback routing.
Read security details →AI agencies
Pain: Need an easy API, but clients demand data privacy.
Solution: Route customer workloads with clearer trust boundaries and verifiable answers.
Read security details →Crypto & Web3
Pain: SaaS centralization is antithetical to the builder ethos.
Solution: Use an open-source, inspectable router with wallet-friendly Web3 sign-in.
Read security details →Dev tools
Pain: Supporting every new API integration is a distraction.
Solution: Support many models explicitly without maintaining every provider SDK.
Read security details →Open-weight leaders, at a fraction of frontier cost.
Qwen, GLM, DeepSeek, Gemma, Kimi and MiniMax — the models most teams actually run — each routed across multiple hosts for price and availability. Frontier models too.
| Model | Maker | Context | Prompt price | Class |
|---|---|---|---|---|
| qwen/qwen3-235b-a22b-2507 | Alibaba | 262k | $0.11 / 1M | Open weights |
| z-ai/glm-5 | Zhipu AI | 205k | $0.66 / 1M | Open weights |
| deepseek/deepseek-v4-flash | DeepSeek | 1M | $0.14 / 1M | Open weights |
| google/gemma-4-31b-it | 262k | $0.14 / 1M | Open weights | |
| moonshotai/kimi-k2-thinking | Moonshot | 262k | $0.66 / 1M | Open weights |
| trustedrouter/eu | EU-focused | Varies | Varies | Regional |
| trustedrouter/auto | Multiple (failover) | Varies | Varies | Attested + ZDR |
| anthropic/claude-sonnet-4.5 | Anthropic | 200k | $3.00 / 1M | Frontier |
A router you can inspect, fork, or run yourself.
For sensitive infrastructure, "trust us" is not enough. TrustedRouter makes its routing layer inspectable, giving technical teams a path to self-host or verify the hosted workload.
Usage-based routing. No subscription required.
Start with prepaid credits, BYOK, or usage-based billing. Teams already spending on LLMs can request migration credits.
⚡ Pay as you go
Wallet accounts can start at $0 credits. Add a card for prepaid credits.
🔑 Bring your own key
BYOK supported for providers where you already have volume discounts.
→ Migration credits
Migration credits by approval for teams spending more than $100/month on LLMs.