Agents and Automation
Agents, at app.devghouse.com/agents, is where you talk to an AI helper about anything in your workspace: draft a blog post, plan next week's content, fix a page on your website, or answer a customer message. Every conversation is saved, so you can close the tab and pick it back up later.
Automation, at app.devghouse.com/routines, is the next step: turn a task you would otherwise repeat by hand into a routine that fires on a schedule, when someone submits a form on your site, or when another system sends it a signal, with nobody watching.
Agents and Automation are always turned on for every workspace, no module to switch on first. A routine that publishes a blog post or checks your SEO still needs that feature (Blog, SEO, and so on) turned on separately to have something to act on.
Who this is for
- Anyone who wants help with a task, right now, without learning a new tool. Type what you want in plain language, the same way you'd ask a coworker.
- Anyone who runs a website, blog, or shop on DevGhouse and wants a recurring job, like a weekly blog post or a daily SEO check, to happen on its own.
- Anyone connecting their own AI assistant (Claude, Codex, or another MCP-compatible tool) to their workspace, to do the same things from outside the app.
What you can do
| Feature | What it does |
|---|---|
| Getting started | Find Agents in your workspace, and see what a first-time, empty screen looks like. |
| Chatting with an agent | Start a one-off conversation, talk to Jarvis by voice, or watch every agent currently working in your context. |
| Agent Library | Save a helper with its own name, personality, and set of tools, so you don't explain the same context twice. |
| Setting up an automation | Build a routine from scratch or a template, and choose what fires it: a schedule, a webhook, or a form. |
| Managing automations | Change a schedule, cap what a run can cost, fire one by hand, and read what each past run actually did. |
| Workflows | A small cast of helpers working one job in stages, useful for a repeatable job like a full blog post, instead of one helper doing everything at once. |
| Connecting your own AI | Connect a Claude, Codex, or Kimi subscription you already pay for, so chats and automations run on it instead of your DevGhouse credits. |
| Troubleshooting | Fixes for the problems people hit most often. |
How it connects to the rest of your workspace
An agent's tools only reach as far as what your workspace has turned on. A helper can only touch your blog if Blog is enabled, only touch a website's content if that website is linked to your project, and so on. When you build an automation, you can link it to a project so it knows which website or codebase it's acting on. Whatever it spends, credits or your own connected subscription, comes out of whichever context (your personal account, or a shared workspace) you're currently in, shown in the switcher at the top of the page.
Run it with an AI assistant
If you connect your own AI assistant (Claude Code, Codex, or another MCP-compatible tool) to your workspace, using a personal MCP token (see MCP Server for how to get one), it can do everything in this guide by calling the platform directly, instead of you clicking through the app.
- Saved helpers:
list_agents,create_agent,update_agent,run_agent,delete_agentmanage one workspace's agents;agent_library_listlists every agent you made, in every workspace. - Conversations:
agent_session_get_conversationreads back what a chat or a run actually said and did;agent_session_deleteremoves one. - Automations:
list_agent_routine_templatesshows the ready-made starting points;create_agent_routine,update_agent_routine,delete_agent_routineset one up, change it, or remove it;run_agent_routine_nowfires one immediately;list_agent_routine_runsandget_agent_routinecheck on it. - Workflows:
workflow_create,workflow_list,workflow_get,workflow_set_seatbuild and inspect a multi-step chart;workflow_run_nowfires one by hand for a one-off job.
Your assistant only ever acts inside the workspace and modules you've given it access to, and nothing runs or gets scheduled by accident.
Credits
Starting a chat, reading a saved helper, or browsing the Library costs nothing. What spends credits is the AI actually doing work, a chat reply, or a routine's run. A routine that runs on DevGhouse's own account shows its cost right on the Routines page (10 per run by default). Connect your own Claude, Codex, or Kimi subscription in Connecting your own AI and a successful run costs 1 credit instead, with a failed run costing nothing.