The AI agent

The AI agent for your local web stack.

It creates sites, edits WordPress, wires n8n workflows, queries MySQL and pushes to GitHub, on your computer, within permissions you set, with a receipt on every change.

Download for free See it ↓
Works with Claude, GPT, Gemini, OpenRouter, Groq, or a free local model
ReDock · AI agent
AWorking on Acme Storedesign & content allowed · database off
▍
The plan · said before any tool runs
Read the site's headerNav.astro✓
Changed the button label to "Download"undo ready✓
Rebuilt the site so it's live12 s✓
Checked the page: the new label renders✓✓
✓Done. Live in 1 m 12 s, every change checkedUndo
Ask for a change, or a question…↑
0 permission levels per site, plain language, not tiers
0 plan, stated before any tool runs
0 terminal commands. It speaks work, not wire
0 providers, bring any key, or run a model locally
Receipts

Every change comes with a paper trail.

The agent states its plan before any tool runs, checks each step after it lands, and folds the run into one line you can reopen later. Receipts read like work ("Changed the button label"), never like wire. A cross-site ledger keeps the history after the chat is gone.

A plan first: numbered steps, checked off liveUndo on the receipt, and in the ledger, days laterFailures never fold away; they explain and offer the fix
Logs & healthThe design laws
ReDock · Activity
A receipt, reopenedWorked 1 m 12 s · 4 steps · every change checkedHide steps ▴
1Read the site's headerNav.astro
2Changed the button label to "Download"Undo
3Rebuilt the site so the change is live12 s
4Checked the live page: the new button shows✓
the same receipt lives in the cross-site ledger: Activity → this site
Permissions

You set the fence. It respects it.

Per site, in words a stranger understands: design & content, management, database, advanced. Deletion is never automatic: the agent asks, every single time. The same fence applies to any AI app you connect over MCP, so there is exactly one permission model to trust.

Per-site switches, two clicks from anywhere"Advanced" asks before each command it runsOne fence for the built-in agent and every connected AI app
MCP serverReDock + Claude Desktop
Site → AI agent → Permissions
What the AI may do onacme.test
Design & content
edit pages, posts, menus and styles
Management
install and update plugins, themes, settings
Database
read and change stored data
Advanced
run commands inside this site's folder · asks first, every time
Deleting things
always asksAsks
Providers

Bring any brain. Or use the free one.

Claude, GPT and Gemini with your own keys, anything via OpenRouter or Groq, or a model from ReDock's catalog running on your computer: free, offline, private. The picker states readiness and cost separately, and a smaller local toolset is declared where you type, before it ever fails.

Keys stay in your system credential store, never on our serversPer-chat model switching at the composerLocal model: no key, no cloud, honest about its limits
ReDock + ClaudeReDock + OpenAIReDock + OpenRouter
Settings → AI
Choose your AI
Claude Sonnet
Anthropic · fast, full toolsetReady
GPT
OpenAI · pay per useAdd a key
Gemini
Google · pay per useAdd a key
OpenRouter
any model, one keyAdd a key
A model on this computer
free · private · works offline · smaller toolset, declared2 GB download
keys live in your system credential store and go only to the provider you picked
It operates everything

One agent, the whole harbor.

The agent drives every part of ReDock; these pages show exactly what it can do in each.

WordPress Pages, plugins, themes, menus: edited like an engineer would, receipt attached. Explore → n8n Start it, open it, share its webhooks; management stays behind its switch. Explore → MySQL Query and browse safely; writes live behind the database permission. Explore → GitHub Pull, commit, push with your device-code token, never your password. Explore → Claude Desktop Prefer chatting there? Drive all of ReDock over MCP with the same fence. Explore → Firecrawl Firecrawl Let the agent read the web when a task needs sources. Explore →
"Tool calls read like work, not wire: 'Changed the button label', never set_post_blocks a1f3." the receipt grammar, from the design laws →
From the blog
Ship AI hosting: what happens when your hosting has an agent attached AI hosting isn't a chatbot bolted onto a dashboard, it's your local dev environment exposed as tools an agent can drive directly. Read → Automation AI scheduling: let an agent plan and queue your week's content AI scheduling pairs an agent that plans your content week with an n8n queue that executes it, self-hosted, free, running on your own Mac. Read → Video Let Claude write your launch video: agent-driven Remotion scenes Describe the video, let an agent write the Remotion scene code, and render an exact MP4 on your Mac, AI video from code, no keys, free with ReDock. Read → Local AI Claude vs local models for agent work: an honest comparison Claude wins on deep reasoning; a local model wins on privacy, cost, and offline. For agent work the honest answer is: use both, matched to the task. Read →
More of the dock
Sites & local HTTPSOne-click sites on real .test HTTPSPublic sharingname.redock.xyz in secondsDatabase browserBrowse and query without a terminalDocumentation studioDocs that capture themselvesVideo studioAI-written scenes, rendered on your computerImage generationPictures made on your own computerLogs & healthErrors that explain themselvesMCP serverDrive ReDock from any AI app
ReDock

Calm ships tonight.

Download ReDock, berth your first site, and watch the AI work, receipts and all. Five minutes, no terminal.

Download for free See pricing
Free · macOS and Windows · your sites stay on your machine