Skip to content
Manifesto
GuideNov 3, 2026·5 min read

Bring your own model: setting up Agentic Space

Connect OpenRouter, Anthropic, Gemini or a local Ollama server, and choose how much the agent may do without asking.


Agentic Space is a chat that can operate Manifesto. Ask "who calls this method?" or "add the INTERNET permission and rebuild", and the agent uses the same tools you do.

Pick a provider

Open Agentic Space → Connect provider:

  • OpenAI-compatible works with any base URL: OpenAI, OpenRouter, Groq and others.
  • Anthropic and Gemini have native support.
  • Local: choose OpenAI compatible and set the base URL to http://localhost:11434/v1 for Ollama, or http://localhost:1234/v1 for LM Studio. Plain HTTP is only allowed to localhost.

Add as many models as you like and switch between them from the composer.

Decide how much it may do

Under the agent settings, choose an approval policy:

PolicyBehavior
Ask for everythingEvery tool call waits for you
Ask before changesLocal reads run freely; edits and remote (MCP) tools wait
AutomaticThe agent runs tools on its own

You can also disable individual tools, for example the terminal.

Extend it

Connect MCP servers over Streamable HTTP or SSE, and import skills as SKILL.md files to teach the agent your own workflows.

Your keys and conversations stay on your device. Requests go only to the provider you choose.

The Agentic Space docs list the settings in detail. Agentic Space is one part of Manifesto's APK editor for Android.