Pi Agent | Apporo

Pi Agent

Home AI workstation on Home Assistant

Read the symptom, not the manual: fixing Pi Agent when it stops working

Read the symptom, not the manual: fixing Pi Agent when it stops working

Every other article explains what to do when things work. This one runs backward: read what is on screen — a red number, a spinner that never stops, a video with no sound — and work back.

Sep 8, 2026

The upgrade that makes all seven of your API keys disappear

The upgrade that makes all seven of your API keys disappear

Pi Agent puts out a release almost every month, and nearly all of them you can simply install. Version 0.13.0 is the exception. It moved every API key field out of the add-on and carried nothing across, so what you meet afterward is 401 errors everywhere. This article is the routine that makes that harmless: what to copy before you press Update, the seven steps, the check afterward, and the way ba

Sep 8, 2026

One Home Assistant backup already carries almost everything Pi Agent holds

One Home Assistant backup already carries almost everything Pi Agent holds

Your conversations, provider keys, Skills and rclone authorization all live in one folder on your Home Assistant host. The backup you may already be running can pick that folder up with everything else — but only if you include the add-on, and only if you have proven once that the archive restores. This article covers what travels, what is left out on purpose, and the five minutes that turn a back

Sep 8, 2026

Getting the finished video out of the container and onto your phone

Getting the finished video out of the container and onto your phone

Part 11 built the video pipeline and Part 12 explained its 720 MB of tooling. The MP4 it produces is real, and your phone cannot reach it. This article connects /data/pi-agent/ to Google Drive with rclone, already inside the add-on. Authorize it once, and every finished video arrives on your phone by itself.

Sep 8, 2026

The 720 MB your first start downloads, and why it never downloads it again

The 720 MB your first start downloads, and why it never downloads it again

The first time you start Pi Agent it goes quiet for several minutes and fills the log with messages. Nothing is wrong. It is fetching the tools the video pipeline needs, once, then writing itself a note so it never has to do it again. This article covers what is in that download, where it lands, when it comes back, and the one switch that brings it back deliberately.

Sep 8, 2026

Turning one written request into a finished, subtitled MP4

Turning one written request into a finished, subtitled MP4

You have something worth showing people — a new robot vacuum, the one-tap Away button on your dashboard, how the recycling gets sorted. Making a watchable video of it is the part that never happens. Pi Agent ships the tools for a video production line and a Skill that drives them: you write a sentence, and a few minutes later there is an MP4 with narration and burned-in subtitles. Here is what tha

Sep 8, 2026

Install a Skill somebody else wrote, then write your own

Install a Skill somebody else wrote, then write your own

Part 9 explained what a Skill is. This one is the doing half: install a published Skill without writing code, learn the naming rule that trips up almost everybody, then write your own — a refrigerator-inventory Skill in a single file. It ends with the failures you are most likely to hit.

Sep 8, 2026

A Skill is a written procedure, not an app you install

A Skill is a written procedure, not an app you install

Every new conversation starts blank. Whatever you explained yesterday is gone, so the model asks again. A Skill is how you stop repeating yourself: one text file, kept in a folder on your own machine, that the model can find and follow on its own. This article is about what that file is and how it gets in front of the model. Installing one and writing one come later.

Sep 8, 2026

Let a model think harder, or hand the conversation to a different one

Let a model think harder, or hand the conversation to a different one

Two skills that belong together. Some models spend extra computation before answering — slower, dearer, sometimes worth it. And you can swap the model mid-conversation without losing a word of it. This article covers when the extra thinking earns its cost, what a switch really does, and the bill that arrives quietly if you switch too often.

Sep 8, 2026

A second provider is four fields, and three of them are easy to get wrong

A second provider is four fields, and three of them are easy to get wrong

You have OpenRouter working. Read this before you add a second key: the three fields that decide whether a route works at all, why one model answers to two different names, and the seven routes the add-on documents. Treat every model ID here as a versioned example and check the provider's live page before saving.

Sep 8, 2026

When one AI provider is enough, and when it stops being enough

When one AI provider is enough, and when it stops being enough

You set up OpenRouter in Part 3 and it works. This article is not here to talk you out of that. It explains the three places where relying on a single route starts to pinch, how to tell whether you have actually hit one of them, and what a second provider would and would not fix.

Sep 8, 2026

The gray block, the white card and the red-and-green lines

The gray block, the white card and the red-and-green lines

Your first conversations had extra furniture around the answer. A collapsed gray block. A white card with a tool name on it. Lines in red and green that look like a correction. None of it is decoration. It is the record of what the agent thought, what it did, and what it wants to change in your files. This article teaches you to read all three, and to know which you can skip.

Sep 8, 2026

1
2