Wardgate Documentation¶
Wardgate is a security gateway for AI agents. It provides credential-isolated API proxying and policy-gated remote command execution (conclaves), with audit logging, approval workflows, and sensitive data filtering.
Quick Start¶
API Gateway -- proxy API calls with credential isolation¶
endpoints:
todoist:
preset: todoist
auth:
credential_env: WARDGATE_CRED_TODOIST_API_KEY
capabilities:
read_data: allow
create_tasks: allow
delete_tasks: deny
Included presets: todoist, github, cloudflare, google-calendar, postmark, sentry, plausible, imap, smtp, pingping
Conclaves -- isolated remote execution for agent commands¶
conclaves:
obsidian:
description: "Obsidian vault (personal notes)"
key_env: WARDGATE_CONCLAVE_OBSIDIAN_KEY
cwd: /data/vault
rules:
- match: { command: "rg" }
action: allow
- match: { command: "tee" }
action: ask
- match: { command: "*" }
action: deny
Documentation¶
Core Concepts¶
- Security Architecture -- How Wardgate protects your credentials and isolates execution
- Policy System -- Writing and configuring rules (allow/deny/ask, rate limits, time ranges)
API Gateway¶
- Presets Reference -- Built-in presets, capabilities, and how to create your own
- Sealed Credentials -- Let agents carry their own encrypted API keys
- Configuration Reference -- All configuration options including sensitive data filtering
Conclaves¶
- Conclaves -- Isolated remote execution environments, policy rules, deployment
- wardgate-cli -- Restricted HTTP client and conclave exec tool for agents
- wardgate-proxy -- Local reverse proxy for agents with their own HTTP clients, or tools
- wardgate-cli AI Skill -- Skill file to teach AI agents how to use wardgate-cli
Operations¶
- Installation -- Pre-built binaries, Docker, building from source
- Deployment Guide -- Docker, Caddy, and production setup
Admin UI & CLI¶
Wardgate includes a web dashboard (/ui/) and CLI for managing approval requests. Configure admin_key_env in your server settings to enable. See the README for an overview.
The dashboard includes:
- Pending -- Requests awaiting approval
- History -- Past approval decisions
- Logs -- Recent request activity with filtering