LLM API for Financial Services and Banks | TrustedRouter
An OpenAI-compatible LLM API banks and fintechs can get approved: hardware-attested confidential compute, zero-data-retention tiers, and docs for DORA reviews.
An LLM API a bank risk committee can verify with one curl command.
Your compliance team blocked the public LLM APIs. GLBA, MNPI, and PCI-scope concerns did the work, and a DORA register now wants specifics about where prompts go. The options left on the table look like a stalled project or an on-prem build. There is a third path. TrustedRouter is an LLM API for banks, insurers, and fintechs that runs the prompt path inside a hardware-attested confidential-compute gateway. The gateway source is open, the image digest is published, and attestation is checkable live, signed by the CPU vendor's root key and bound to the TLS session so it cannot be replayed.
TrustedRouter does not store prompt or output content by default. Every route carries a privacy tier: Zero-Data-Retention providers contractually keep nothing; TEE routes keep the prompt sealed end to end, so even the serving provider cannot read it. The API is OpenAI-compatible, so migration is a base_url change. Keep your SDK, your model ids, your code. Privacy you can check beats privacy you're promised.
NONCE=$(openssl rand -hex 16)
curl "https://api.trustedrouter.com/attestation?nonce=$NONCE"
# Returns a JWT signed by the CPU vendor's root key.
# image_digest matches the published open-source build.
# Satisfied? Point your existing SDK at:
# https://api.trustedrouter.com/v1A sub-processor story that survives bank review
Fintech vendors selling into banks get the paperwork up front: DPA, BAA, public sub-processor list, plus SOC 2 readiness and HIPAA readiness documentation. None of it claims a certification. It is a verifiable architecture plus readiness documentation your compliance team can assess.
Register entries and an exit strategy
An ICT third-party register wants named sub-processors, documented data handling, and a credible exit plan. Exit is structural here: OpenAI-compatible means leaving is a base_url change, and 220+ model routes across 30+ providers with automatic fallback answer the concentration question. An EU gateway provides EU-focused routing. Informational, not legal advice.
Tiers your data policy can map to
Choose the tier per route. Open runs on any attested provider. Zero-Data-Retention keeps nothing by contract. TEE seals the prompt end to end. Route MNPI, banking data, and customer records through TEE, and check measured latency from real routed requests before you commit.
Between a blocked API and a rack of GPUs.
On-prem gives banks and insurers control, and it costs a platform team: hardware, model ops, patching, capacity planning. A public API is fast and fails the review. Attested confidential inference sits between them: API economics with cryptographic proof of where the prompt runs. Attestation beats a privacy policy.
It holds up under load. A document-processing company scaled from zero to over a billion tokens per week through the attested gateway within three weeks of onboarding. Start in the playground, pick a route at /choose, and see per-model pricing. Verify, don't trust.
Questions
Is TrustedRouter SOC 2 certified or HIPAA compliant?
No. TrustedRouter publishes SOC 2 readiness and HIPAA readiness documentation, and those are readiness documents rather than audit reports or certifications. The offer is a verifiable architecture plus readiness documentation your compliance team can assess, alongside a DPA, a BAA, and a public sub-processor list your procurement process can reference.
How does TrustedRouter fit a DORA third-party risk review?
TrustedRouter provides the material an ICT register asks for: a public sub-processor list, a DPA, an open-source gateway with a published image digest, and live attestation you can check yourself. The exit strategy is structural, since the OpenAI-compatible API makes migration a base_url change. An EU gateway offers EU-focused routing. This is informational, not legal advice.
Does TrustedRouter store our prompts or outputs?
TrustedRouter does not store prompt or output content by default. Each route carries a privacy tier: Open runs on any attested provider, Zero-Data-Retention providers contractually keep nothing, and TEE keeps the prompt sealed so even the serving provider cannot read it. The prompt path runs inside a hardware-attested confidential-compute gateway you can verify live.
Do banks have to run LLMs on-prem to protect customer data?
On-prem is one answer, and it costs hardware, a platform team, and ongoing model ops. Attested confidential inference is the middle path: prompts run inside a hardware-attested TEE, the gateway source is open, the image digest is published, and attestation binds to the live TLS session so it cannot be replayed. You can check all of this with a curl command before sending real data.