Every model, one API
80+ open-weight models behind one OpenAI-compatible endpoint. Change two lines and keep your SDK.
Browse modelsSent as Morse
Providers see Morse as the customer. Your account, your card, and your IP address stop at Morse.
See where a prompt goes
Change two lines. Point the OpenAI SDK at Morse and use your Morse key. Everything else stays the same.
Works with what you already use.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.openai.com/v1", (removed)
api_key=os.environ["OPENAI_API_KEY"], (removed)
base_url="https://inference.trymorse.ai/v1", (added)
api_key=os.environ["MORSE_API_KEY"], (added)
)
reply = client.chat.completions.create(
model="moonshotai/kimi-k3.1-pro",
messages=[{"role": "user", "content": "Hello"}],
)
Set the privacy of every request. From providers that keep nothing to sealed hardware no one can see into, with one field in your request.
Your app asks for the TEE tier. Morse can't read your prompt, and sends the request to GLM 5.3 on Phala, attested hardware.
Most APIs decrypt at the edge. Morse decrypts only inside sealed hardware.
HTTPS protects a request only until the first machine holding the key. Anyone with access to that machine, or to the systems behind it, can read your prompts.
Conventional API
Decrypted at the edge
TLS terminates at the edge. A load balancer or CDN holds the key and decrypts every request.
Every system behind it can read your prompt. Routing, logging, and billing systems receive your prompt as plain text.
You have to trust the provider. Their privacy policy is a promise, and you have no way to check it.
Morse
Decrypted only in the enclave
TLS terminates inside the enclave. An enclave generates the key at boot, and the key never leaves enclave hardware. The network between you and the enclave carries only ciphertext.
Nothing else Morse runs can read your prompt. The enclave encrypts each request again before it leaves. Billing sees your key, the time, the model, the provider, and the token count, but not the content.
You can check it. AWS Nitro signs an attestation that names the key. Every certificate is in public logs, and every boot is recorded in a public log.
Check it yourself
One command checks the key against the enclave's attestation.
How verification works npx @askmorse/verify inference.trymorse.aiEvery request, optimized. The savings come back to you.
One price per model.Every model's price is on the models page, before you send a request.
Adaptive routing.Each request goes to the best-priced vetted provider at the privacy tier you pick, never a less private one.
Dynamic caching.Repeated context is served from a warm cache, often at a tenth of the input price. We never read your prompts to do it.
Up to 50% back.Up to 50% of every dollar saved is refunded to your balance as credit after each UTC hour closes, for that hour's requests. We only do well when you save.
DeepSeek V4.1 Flash 1017
$0.14You paid
+$0.05Back to you
Refunded back to you this sessionClick a request or a step to see its breakdown.
+$0.23How Morse compares. Against other gateways and inference providers.
 Morse |   |   |   | |
|---|---|---|---|---|
| Privacy | ||||
| Prompts hidden from the gateway | Yes, Can't see your prompts | No, Sees your prompts, even with zero data retention | No, Sees your prompts, even with zero data retention | Partly |
| Prompts hidden from the model provider | Yes, On TEE and E2EE models | No, The provider sees your prompts | Yes, On secure-hardware models | No, The provider sees your prompts |
| Your identity hidden from the provider | Yes, The provider only sees Morse | Partly, Shared by default | Yes, The provider only sees Venice | No, You use your own accounts |
| Trust | ||||
| Open-source code | Yes, Its enclave | No | No | Yes |
| You can verify the code it runs | Yes | No | No, Secure-hardware models only | No |
| Models on secure hardware | Yes, TEE and E2EE | No | Yes | No |
| Transparency | ||||
| Lists every provider publicly | Yes | Yes, On each model's page | No, Secure-hardware partners only | Partly, You pick your own |
| Shows who served each request | Yes, In a signed receipt | Yes | No | Yes, In your logs |
| Models | ||||
| One key for every model | Partly, Open chat models | Yes | Yes | Partly, A key for each provider |
| Newest uncensored models | Partly, In the Morse app | No, Only a few | Yes | Partly, Depends on your providers |
From each company's public docs, November 2026.
Common questions. About privacy, providers, and price.
How is this different from OpenRouter's zero-retention mode?
OpenRouter's zero data retention covers the providers OpenRouter sends your prompts to, not OpenRouter itself: OpenRouter's servers read every prompt in plain text, and their code is closed. Morse's enclave runs open-source code you can check, and a request your client encrypts for Morse's enclave can be decrypted only inside that enclave.Verify the enclave
How does Morse make money?
We pay providers wholesale prices and charge you more than we pay. Teams also pay for extras like dedicated enclaves and audit reports. We never sell your data, and never have any to sell.
Why should I trust the providers?
On Private, you rely on contracts: every provider has signed a zero data retention agreement to store none of your prompts and never train on them, and every provider is on our public list. On TEE and E2EE, you don't have to trust the provider at all: the model runs in the model provider's enclave, hardware even the provider can't see into.Compare the three tiers
Why not go to the provider directly?
You can, and for one open model it may be cheaper. With Morse, the provider never learns who you are, one key reaches 80+ models under zero data retention agreements, and one set of paperwork covers them all.
What doesn't Morse protect against?
On the Private tier, the model provider reads your prompt in plain text, and you rely on the provider's agreement to store none of your prompts.Compare the three tiers