Untrusted by Design
Every agent is a container โ reproducible, inspectable, diffable, and killable. Purpose is bind-mounted read-only. Survives full container compromise.
Infrastructure-layer governance for AI agent containers. The layer below the framework, where deployment meets governance.

Clawfile โAn extended Dockerfile. Any valid Dockerfile is a valid Clawfile.
FROM openclaw:latest
CLAW_TYPE openclaw
AGENT AGENTS.md
MODEL primary openrouter/anthropic/claude-sonnet-4
CLLAMA passthrough
HANDLE discord
INVOKE 15 8 * * 1-5 pre-market
SURFACE service://trading-api
SURFACE volume://shared-research read-writeclaw-pod.yml โAn extended docker-compose.yml. Services inherit pod-level defaults.
x-claw:
pod: trading-desk
master: octopus
cllama-defaults:
proxy: [passthrough]
env:
OPENROUTER_API_KEY: "${OPENROUTER_API_KEY}"
surfaces-defaults:
- "service://trading-api"
- "volume://shared-research read-write"
services:
tiverton:
image: trading-desk-tiverton:latest
build:
context: ./agents/tiverton
x-claw:
agent: ./agents/tiverton/AGENTS.md
handles:
discord:
id: "${TIVERTON_DISCORD_ID}"
username: "tiverton"curl -sSL https://raw.githubusercontent.com/mostlydev/clawdapus/master/install.sh | sh
git clone https://github.com/mostlydev/clawdapus.git
cd clawdapus/examples/quickstart
cp .env.example .env # add your keys
claw pull # pinned runtime infra + registry services
claw build # local build: services
claw up -d # compile + launch
claw health # โ all healthy
claw down # tear down when you're doneThe everyday operator loop is claw pull, claw build, claw up, claw down.
claw.describe, compiled by claw up, and orchestrated by cllama. Infrastructure recalls derived context before each inference turn and retains after each response โ automatically, without the agent asking.