Dokumentation
Dokumentation zu Agent host
Erstell Agenten ohne Code: Anweisungen, Fähigkeiten, Wissen, einen Zeitplan und Werkzeuge über MCP-Connectoren. Unterhalte dich in Echtzeit mit ihnen. Alles Riskante wartet in einem Freigabe-Posteingang, bis eine Person Ja sagt.
Die Anleitungen sind bisher auf Englisch.
Hier beginnen Übersicht auf Englisch
Start
Getting started
Your first agent in Agent host: pick a template, write what it is for, deploy it, chat with it and approve a risky action. About twenty minutes.
Plans and limits
The Agent host plans: Pro €19 with 1 agent on your own model key, Premium €39 with 3 agents, Max €79 with 10 agents. Every plan runs on your own model key.
Define
Define an agent
What goes into an Agent host agent definition: description, instructions, persona, skills, knowledge, model tier and limits, validated on save.
Templates
The built-in Agent host templates: inbox triage, research digest and a project reporter, and what each is set up to do and not do.
Skills and knowledge
Skills and knowledge in an Agent host agent are plain text that goes into its prompt: how to write them and what not to put in.
Tools
Connectors
How to give an Agent host agent tools through MCP connectors: any MCP server by https address, safe tools, and where secrets are kept.
Web tools and the webhook
The built-in tools of an Agent host agent: web fetch and web search, which are safe, and the outbound webhook, which always asks first.
Run
Schedules
How an Agent host agent wakes on a schedule: the interval, the prompt, the shortest interval per plan, the digest of the last run and the daily tool-call limit.
Chat
Chat with a deployed Agent host agent in real time: how the connection is made, what you see, and limits on message length.
Versions
Versions in Agent host: every save is immutable, deploy activates one and pushes it to the agent, roll back activates an earlier one.
Models and keys
Which model an Agent host agent uses: your own key on every plan, and model tiers haiku, sonnet and opus.
Usage
Where to see what your plan holds and what is used: agents and the daily tool-call limit.
Logs and costs
The logs of an Agent host agent: runs per day with their steps.
Folders and tags
Organise Agent host agents with folders, tags, favourites and archiving, and find them by text, status or tag.
Approve
The risk rule
The one rule that decides what an Agent host agent may do alone and what waits for a person: safe tools, risky tools and what is not offered.
Approvals
How approvals work in Agent host: a risky call is suspended, appears in the inbox, and a person of the workspace approves or denies it. Your AI cannot.
Connect
Prompt Verse
Link an Agent host agent to a definition authored in Prompt Verse: pin a version or follow the latest, and optionally deploy on update.
Use Agent host with your AI
Connect Claude, ChatGPT, Grok or any MCP assistant straight to Agent host: its own MCP address, sign-in or API key, 1 tool and 6 actions.
Reference
Help
Your data
What Agent host stores: definitions as versions, chats, runs, approvals and usage, where connector secrets live, and what a model provider sees.
Troubleshooting
Fixes for common Agent host problems: an agent that will not chat, an approval that never ends, a schedule that is refused, a limit reached, no model key.
Glossary
The words used by Agent host in plain language: agent, definition, version, connector, safe tool, risky tool, approval, digest and tier.