From 76dab6ff92afa9eec9263b128b33d13fb54f97ed Mon Sep 17 00:00:00 2001
From: DavidG
Let any AI agent use the web: reads of trusted hosts pass,
+
Let an AI agent running in lea use the web: reads of trusted hosts pass,
anything that sends data or reaches an unknown host waits for you.
@@ -32,10 +32,17 @@ anything that sends data or reaches an unknown host waits for you.
lockhttp is a Lock of the [Lock Family](https://github.com/DavidGodefroid/lockfamily), harness security for AI agents. It is the proxy every outbound HTTP(S) -request of your agent goes through. Reads of hosts you trust pass; anything -else waits for **your** approval on +request of an agent running in [lea](https://github.com/DavidGodefroid/lea) +goes through. Reads of hosts you trust pass; anything else waits for +**your** approval on [lockhub](https://github.com/DavidGodefroid/lockhub), where the agent cannot reach it. +> [!IMPORTANT] +> **The agent must run inside lea.** lockhttp guards only the containers lea +> starts with `LEA_LOCKHTTP=1`, whose only way out is lea's proxy. An agent +> you start directly on your machine (outside lea) does not go through the +> gate: lockhttp sees none of its requests and protects nothing there. + > [!NOTE] > **0.x**: the policy format may still change between minor versions. Built on [lockkit](https://github.com/DavidGodefroid/lockkit) v0.14.1. > Command names are fixed; flags may still move. Design decisions: [docs/decisions.md](docs/decisions.md). @@ -113,9 +120,9 @@ commands and the setup by hand: [docs/install.md](docs/install.md). 4. Ask your agent to read the docs of a trusted site: it goes through. Ask it to post somewhere: the request waits for your `y`. -lockhttp is a proxy, so it works with any agent and any HTTP client inside -lea's containers (lea runs Claude Code today). The contract with lea: -[docs/lea.md](docs/lea.md). +lockhttp is a proxy, so inside lea's containers it works with any agent and +any HTTP client (lea runs Claude Code today). Outside lea it is not on the +path at all. The contract with lea: [docs/lea.md](docs/lea.md). ## Highlights @@ -234,7 +241,7 @@ The site, [lockhttp.com](https://lockhttp.com), is built from [`site/`](site/). | [locksql](https://github.com/DavidGodefroid/locksql) | SQL databases: the agent queries, PII comes back masked; connections and writes wait for you. | | [lockmq](https://github.com/DavidGodefroid/lockmq) | RabbitMQ: masked queue inspection; connections, peeks and unmasked reads wait for you. | | [lockfile](https://github.com/DavidGodefroid/lockfile) | Files (S3, SFTP, disk): the agent asks yes/no questions about files it never reads. | -| [lockhttp](https://github.com/DavidGodefroid/lockhttp) | Outbound HTTP(S): reads of trusted hosts pass; anything that sends data or reaches an unknown host waits for you. | +| [lockhttp](https://github.com/DavidGodefroid/lockhttp) | Outbound HTTP(S) of agents in lea: reads of trusted hosts pass; anything that sends data or reaches an unknown host waits for you. | | [locklog](https://github.com/DavidGodefroid/locklog) | Logs: masked, bounded log search; the agent reads the logs, not the people in them. | | [lockssh](https://github.com/DavidGodefroid/lockssh) | Commands over SSH: allow-listed commands run at once, the rest waits for you; the key never leaves the console. | | [lockhub](https://github.com/DavidGodefroid/lockhub) | The approval hub and its `lox` CLI: `lox run` approves outside the agent's harness and never sees values in clear. | diff --git a/docs/security-model.md b/docs/security-model.md index 6a2694e..ad8e054 100644 --- a/docs/security-model.md +++ b/docs/security-model.md @@ -1,6 +1,7 @@ # LockHTTP security model -LockHTTP puts every outbound HTTP(S) request of an AI agent through a gate +LockHTTP puts every outbound HTTP(S) request of an AI agent running in +[lea](https://github.com/DavidGodefroid/lea) through a gate that decides it with a policy only a human loosens, and holds what needs a human until that human answers in a console the agent cannot reach. This document says what it protects, how, and where it stops. It describes @@ -34,7 +35,10 @@ image, through the launcher. dependency it runs), and hostile web content that steers it. - **Out of scope:** a compromised host or console account, root, any account that may run `docker` on the host (see - [Docker](#what-docker-does-and-does-not-separate)), Anthropic itself, an attacker on the network upstream of the proxy container. + [Docker](#what-docker-does-and-does-not-separate)), Anthropic itself, an attacker on the network upstream of the proxy container, + and any agent that does not run inside lea: an agent started directly on + the host never goes through the gate, so LockHTTP sees and protects none + of its requests. ## Boundaries diff --git a/site/faq/index.html b/site/faq/index.html index 886d75b..a763203 100644 --- a/site/faq/index.html +++ b/site/faq/index.html @@ -65,8 +65,8 @@Any agent and any HTTP client: lockhttp is a proxy, not an agent plugin. It sees requests, whatever tool sent them.
-The agent must run in lea, which runs Claude Code in containers today.
+Any agent and any HTTP client inside lea: lockhttp is a proxy, not an agent plugin. It sees the requests of lea’s containers, whatever tool sent them.
+The agent must run in lea, which runs Claude Code in containers today. An agent run directly on your machine, outside lea, does not go through lockhttp: the gate never sees its requests and does not protect it.
v0.9A Lock of the Lock Family · open source · Apache 2.0
Every outbound HTTP(S) request of your AI agent goes through lockhttp. Reads of trusted hosts pass; anything that sends data or reaches an unknown host waits for you.
+Every outbound HTTP(S) request of your AI agent running in lea goes through lockhttp. Reads of trusted hosts pass; anything that sends data or reaches an unknown host waits for you.
brew trust davidgodefroid/tap
@@ -161,7 +161,7 @@ Highlights
Quick start
- lockhttp is a proxy, so it works with any agent and any HTTP client. It needs lea, which runs agents (Claude Code) in containers whose only way out is its proxy.
+ lockhttp is a proxy, so it works with any agent and any HTTP client, as long as they run inside lea, which runs agents (Claude Code) in containers whose only way out is its proxy. An agent started directly on your machine, outside lea, does not go through lockhttp, and lockhttp does not protect it.
Install with Homebrew (the box above). Docker must be running.
In lea’s projects directory, lox add lockhttp asks for the lock’s name, writes the policy and starts its console.
diff --git a/site/security/index.html b/site/security/index.html
index 76a795f..c11ee52 100644
--- a/site/security/index.html
+++ b/site/security/index.html
@@ -77,7 +77,7 @@ What it protects
Assets: the project’s and the host’s data, credentials, and actions taken online in your name.
Attackers: the agent itself (prompt injection, drift) and hostile web content.
- Out of scope: a compromised host or console account, any account that may run docker on the host, the model provider itself (Anthropic), a network attacker upstream.
+ Out of scope: a compromised host or console account, any account that may run docker on the host, the model provider itself (Anthropic), a network attacker upstream, and any agent not run inside lea: an agent started directly on the host never goes through the gate.