Raven — My Self‑Hosted Media Server AI

[sam9s]

# Meet Raven: My Fully Automated, Telegram-Controlled Media Server AI
Hey everyone! I wanted to share a passion project I’ve been building over the last few months. Like many of you, I have a massive self-hosted media server stack (Jellyfin, Navidrome, Kavita, qBittorrent, etc.).
But managing it all—searching for torrents, fixing broken subtitles, hunting down high-quality FLAC audio, and moving files to the right folders—was becoming a part-time job.
I didn't just want a media server anymore. I wanted a **media assistant**. So, I built **Raven**.
---
## What is Raven?
Raven is an AI operator that lives 24/7 on my VPS. I don't use SSH or web dashboards to manage my media anymore. **I just text Raven on Telegram, and it does the work.**
It isn't a chatbot or a "wrapper" around ChatGPT that just gives me advice. It is a genuine automation engine wired directly into my tools.
I literally just text it things like:
> *"Download the movie Death Sentence (2007) and get English subs."*
> *"Grab the album 'Rumours' in FLAC from Soulseek."*
> *"Pull this YouTube track into my Hindi music folder as FLAC."*
> *"Why is my Jellyfin movies folder empty?"*
Raven reads the message, executes the backend commands, monitors the download progress, renames the files perfectly, sorts them into my Google Drive mount, and replies on Telegram when it's done.
---
## How it Works (Under the Hood)
Raven is powered by **OpenClaw**.
For those who don't know, **OpenClaw** is a brilliant, open-source AI framework that lets you build persistent AI agents that can securely execute CLI commands, run scripts, and manage your local or cloud servers. (You can check out OpenClaw and set it up for yourself here: *https://openclaw.ai/*).
Instead of interacting with dozens of different web UIs, I gave OpenClaw specific "skills" (Python scripts) to talk to my APIs:
**Movies & TV (Jackett & qBittorrent)**
Raven searches private trackers, picks the best release based on file size and seeders, sends it to qBittorrent, and categorizes it automatically (e.g., Hollywood vs. Bollywood).
**Smart Subtitles (SubDL)**
If an obscure movie doesn't have subtitles, I just ask Raven. It searches SubDL, matches the exact release hash, and drops the `.srt` file perfectly named next to the `.mkv`.
**Music (Soulseek & YouTube)**
Finding good FLAC files is usually a headache. Raven searches the Soulseek network for me. If a peer cancels the transfer or stalls out, Raven automatically detects the failure, searches for a fallback peer, and tries again until the album is safely in my Navidrome folder. It even automatically tags tracks and pulls high-quality cover art.
**Books & Comics (Kavita)**
Just text the name of a book, and Raven pulls the EPUB straight into my Kavita reader.
---
## Why I Built It This Way
The biggest problem with typical "AI automation" is that it pretends everything is fine even when things break.
I designed Raven to be brutally honest. If a YouTube download gets blocked by a bot-check, or if a Soulseek user goes offline, Raven messages me on Telegram: *"Hey, the download stalled. Trying a different peer for the remaining 3 tracks."*
It knows my exact folder paths, it knows my tools, and it acts exactly like a human librarian managing the archives.
## Build Your Own!
If you are comfortable with a Linux VPS, Docker containers, and a basic media stack, you can absolutely build this. All you need is **OpenClaw**, an OpenAI/Anthropic API key, Telegram, completely replacing your existing *arr/media stack!
Just wanted to share this to hopefully inspire some of you. The era of clicking through 5 different web UIs just to download a movie is over. Just text your server!
Let me know if you guys have any questions about how the Python scripts hook into the torrent clients or Soulseek. Happy to share how the plumbing works!

A detailed write-up is here if anyone is interested in reading it properly.

# Raven — My Self‑Ho... - MarkdownBin

# Raven — My Self‑Hosted Media Server AI I’ve been building a self‑hosted media setup with an AI ... Share your markdown with the world

mdbin.sivaramp.com

Below is how a typical conversation with my Raven looks.

https://ibb.co/svGtwN89

Below is how the dashboard looks



Below is how a typical conversation with my Raven looks.

https://ibb.co/svGtwN89

 

[sam9s]

Hey guys, thanks for the likes! I wanted to share a quick update on where the project is heading next.
After successfully wiring Raven up to my media stack, I’m now giving it full control over my **Google Ecosystem** (Gmail, Drive, Calendar, and Sheets).
Instead of manually clearing out my inbox or tracking finances, I can just text Raven things like:
> *"Trash all promotional emails from the last 6 months."*
> *"Analyze my inbox and tell me who the top 5 spammers are, then build a filter to auto-delete them."*
> *"Find all those huge random files hoarding space in my Google Drive and clear them."*
It's completely transforming how I handle my personal digital clutter—it's basically a self-hosted personal assistant now. I'll share more details once the code is finalized, but let me know if anyone has built something similar!

 

[reignofchaos]

Isn't openclaw blocked by anthropic now? Also its a huge security hazard to give it the API key.

 

[sam9s]

Isn't openclaw blocked by anthropic now? Also its a huge security hazard to give it the API key.

No, nothing is blocked, and Anthropic just wanted to change its name. Anyhow, the creator of OpenClaw has been hired by Sam Altman, OpenAI, so all the security issues are now being taken care of with the new versions rolling out directly from within OpenAI's lab.

 

[arunlouie]

Hey @sam9s , Glad to see Openclaw reached HFV also. I always liked your HTPC posts, now GenAI, Nice. Very good use case of OPenclaw for AV enthusiast.

I also read Anthorpic started blocking Openclaw just to preserve user token to be wasted. Also is it safer for sharing personal data? I would suggest not to give personal Mailbox access yet. Should try NemoClaw form Nvidia for better guardrails (just from keynote I am saying, yet explore). Have you done any gateways for checking those

Now please share you setup, VPS is hosted cloud or local machine (specs?). You have connected to which LLM?

 

[sam9s]

Hey @sam9s , Glad to see Openclaw reached HFV also. I always liked your HTPC posts, now GenAI, Nice. Very good use case of OPenclaw for AV enthusiast.

I also read Anthorpic started blocking Openclaw just to preserve user token to be wasted. Also is it safer for sharing personal data? I would suggest not to give personal Mailbox access yet. Should try NemoClaw form Nvidia for better guardrails (just from keynote I am saying, yet explore). Have you done any gateways for checking those

Now please share you setup, VPS is hosted cloud or local machine (specs?). You have connected to which LLM?

Yet to try nemoclaw... these are so many claw variants now ... but since Peter (OC Creater) joined open AI ... all my reservations were addressed .. I am constantly following him on Twitter so I know what is happening .....
coming to Anthropic banning ... that was stupid for them, they underestimated the power if OC and Gen AI used for personal tasks, that's what people are using it most for, irrespective of initial security issues.
Anthropic banned to use the OAuth for claude account. Instead, forced people to use their API for OC ....Big mistake.. Open AI on the other hand, since peter already was working for Open AI .. allowed all user to use OAuth AND API both .. which was a kick Ass move ...
This will also probably answer your next question, which LLM I am using .. I started with Sonnet 4.6, then shifted to KIMI 2.5 , and now finally settled with Open AI OAuth.. with gpt 5.4 (which has kicked some serious ass of opus 4.6 in terms of coding and general understanding)

I will soon start to share my setup slowly, just need to see how many are interested to actually build something like this .... and yes I am using VPS on Hostinger, which is where all my media apps are installed and OC as well obviously ...

coming to "giving access" to personal mailbox, well its perfectly fine IF you really know how to setup security guardrails and you know how to configure the access and "skills" used for this purpose .... I have been experimenting with OC since the day it was launched and only after experimenting with it thoroughly (in and out) I have now configured it to control my Google Echosystem along with yes my media server.

media server was the primary need, google came just as an extra added feature ... I have other plans as well ... with OC, like Home automation, controlling streaming service like Spotify..etc etc .. .all will be discussed here, if members show interest ...

 

[dvraghavan]

great to see everyday use cases of OC. Kudos

 

[arunlouie]

Yet to try nemoclaw... these are so many claw variants now ... but since Peter (OC Creater) joined open AI ... all my reservations were addressed .. I am constantly following him on Twitter so I know what is happening .....
coming to Anthropic banning ... that was stupid for them, they underestimated the power if OC and Gen AI used for personal tasks, that's what people are using it most for, irrespective of initial security issues.
Anthropic banned to use the OAuth for claude account. Instead, forced people to use their API for OC ....Big mistake.. Open AI on the other hand, since peter already was working for Open AI .. allowed all user to use OAuth AND API both .. which was a kick Ass move ...
This will also probably answer your next question, which LLM I am using .. I started with Sonnet 4.6, then shifted to KIMI 2.5 , and now finally settled with Open AI OAuth.. with gpt 5.4 (which has kicked some serious ass of opus 4.6 in terms of coding and general understanding)

I will soon start to share my setup slowly, just need to see how many are interested to actually build something like this .... and yes I am using VPS on Hostinger, which is where all my media apps are installed and OC as well obviously ...

coming to "giving access" to personal mailbox, well its perfectly fine IF you really know how to setup security guardrails and you know how to configure the access and "skills" used for this purpose .... I have been experimenting with OC since the day it was launched and only after experimenting with it thoroughly (in and out) I have now configured it to control my Google Echosystem along with yes my media server.

media server was the primary need, google came just as an extra added feature ... I have other plans as well ... with OC, like Home automation, controlling streaming service like Spotify..etc etc .. .all will be discussed here, if members show interest ...

As usual, you go deep in this too. Few quesitons,

1) What is the size of vps and cost you paying (I assume you have it ON 24x7 / our day time)
2) What are the gaurdrails you have currently setup
3) How you local media is accessed/ local server is taking actions from OC (like say downloading a file) ?

Keen to know more, keep sharing.