From 143f2d7162b782bf97fe08ce5aa7a59140ec997b Mon Sep 17 00:00:00 2001 From: Marketen Date: Thu, 10 Sep 2026 09:08:46 +0200 Subject: [PATCH] Don't claim only the TEE reads prompts MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit That holds only for private/ models. With any other model the Gateway, inside the TEE, passes the prompt on to the model's provider, who reads it. The docs now say what is always true — encrypted all the way to Nexus, which runs in a TEE, and nobody in between can read it — and the private/ tip adds that other models are run by their provider, which sees the prompt. Co-Authored-By: Claude Opus 5 --- README.md | 9 +++++---- dappnode_package.json | 2 +- getting-started.md | 9 +++++---- 3 files changed, 11 insertions(+), 9 deletions(-) diff --git a/README.md b/README.md index 3580a2f..80a2b2e 100644 --- a/README.md +++ b/README.md @@ -1,8 +1,8 @@ # Nexus Proxy -Encrypts your Nexus prompts so only the **TEE** — the trusted execution -environment Nexus runs inside — can read them. Not us, not Dappnode, nobody in -between. It verifies the TEE automatically on every connection. +Keeps your prompts encrypted all the way to Nexus, which runs inside a **TEE** +(trusted execution environment). Nobody in between can read them — not us, not +Dappnode. It verifies the TEE automatically on every connection. ## Use it @@ -16,7 +16,8 @@ switch that does this for you — Hermes Agent is one. For the whole path, pick a model whose id starts with **`private/`**. Those run inside a TEE too, so your prompt stays encrypted from this proxy to Nexus and -from Nexus to the model. +from Nexus to the model. Other models are run by their provider, which sees the +prompt to answer it. ## Check it yourself diff --git a/dappnode_package.json b/dappnode_package.json index 99f208e..375a9a7 100644 --- a/dappnode_package.json +++ b/dappnode_package.json @@ -5,7 +5,7 @@ "upstreamRepo": "dappnode/dappnode-nexus-sdk", "upstreamArg": "UPSTREAM_VERSION", "shortDescription": "Keep your Nexus prompts private, and verify it yourself", - "description": "Encrypts your Nexus prompts so only the TEE (trusted execution environment) that Nexus runs inside can read them — not us, not Dappnode, nobody in between. It verifies the TEE automatically on every connection.\n\nYou do not have to take our word for it: the verification page shows what was checked, right now, and lets you download the raw proof to verify independently. It never shows your prompts.\n\nPoint any OpenAI-compatible app on this Dappnode at http://nexus-proxy.dappnode.private:3301/v1 with your normal Nexus API key. Same key, same models, same prices. Some apps, such as Hermes Agent, have a private mode switch that does this for you.\n\nFor the whole path, pick a model whose id starts with private/ — those run inside a TEE too, so your prompt stays encrypted from the proxy to Nexus and from Nexus to the model.\n\nAuto Router (nexus/auto) and PII masking do not work through the proxy.", + "description": "Keeps your prompts encrypted all the way to Nexus, which runs inside a TEE (trusted execution environment). Nobody in between can read them — not us, not Dappnode. It verifies the TEE automatically on every connection.\n\nYou do not have to take our word for it: the verification page shows what was checked, right now, and lets you download the raw proof to verify independently. It never shows your prompts.\n\nPoint any OpenAI-compatible app on this Dappnode at http://nexus-proxy.dappnode.private:3301/v1 with your normal Nexus API key. Same key, same models, same prices. Some apps, such as Hermes Agent, have a private mode switch that does this for you.\n\nFor the whole path, pick a model whose id starts with private/ — those run inside a TEE too, so your prompt stays encrypted from the proxy to Nexus and from Nexus to the model. Other models are run by their provider, which sees the prompt to answer it.\n\nAuto Router (nexus/auto) and PII masking do not work through the proxy.", "type": "dncore", "backup": [ { diff --git a/getting-started.md b/getting-started.md index 111128e..b003e7b 100644 --- a/getting-started.md +++ b/getting-started.md @@ -1,8 +1,8 @@ # Nexus Proxy -Encrypts your Nexus prompts so only the **TEE** — the trusted execution -environment Nexus runs inside — can read them. Not us, not Dappnode, nobody in -between. +Keeps your prompts encrypted all the way to Nexus, which runs inside a **TEE** +(trusted execution environment). Nobody in between can read them — not us, not +Dappnode. It verifies the TEE automatically on every connection. @@ -28,7 +28,8 @@ that does this for you — Hermes Agent is one. Pick a model whose id starts with **`private/`**. Those run inside a TEE too, so your prompt stays encrypted the whole way — from this proxy to Nexus, and -from Nexus to the model. +from Nexus to the model. Other models are run by their provider, which sees the prompt to +answer it. ## Good to know