How-to guides

Task-oriented recipes. Each one solves a single, concrete problem.

Set up and run

Install the desktop app: download a prebuilt installer and run the app, no build required.
Install and run the runtime: build from a checkout and get a daemon (and the desktop dev app) running an agent.
Get the most from local inference: how the embedded llama-server engine serves local GGUF models, and how to feed it.

Write a tool: a Python function an agent calls, declared with @tool and granted in the manifest of the package.
Write a workflow: steps you write in Python, each handing work to an agent, to another workflow or to the user.
Make an agent plan first: have an agent write the steps of a task before it acts.
Test your tools and workflows: unit-test them against a mock context, then evaluate whole tasks against a live runtime.
Package and distribute: build, sign, install and share a .apo archive.

Integrate via the driving contract: control the runtime from a host application over the HTTP API.
Embed via federation: expose your product as MCP tools and let an agent write back through your REST API.
Set up a Google OAuth client: register your own client in the Google Cloud console, click by click, so the Gmail, Calendar and Drive connector works. Microsoft 365 needs no equivalent.
Keep a human in the loop: stop a workflow on a question, and know what asks for an approval on its own.
Audit and verify a run: read the trail and verify the journal.
Deploy in production: run Apollia as a managed service with TCP, authentication, and TLS.