agentos.guide › blog
To get hermes agent video analysis running in your Agent OS, give one profile the opt-in video toolset and a video-capable auxiliary model, use the bundled youtube-content skill for YouTube links, and add the video-xray plugin on a Windows Desktop machine for long recordings.
This guide walks through each of those steps in order, covers what each route can and cannot do, and finishes with three prompts that run the work for you.
Every setting comes from the Hermes tools reference, the Hermes source and the plugin's own catalog entry.
If you'd like help deciding which profile should own video work, book a free SEO strategy session with my team.
Your Agent OS already runs Hermes, a gateway and several profiles side by side.
Video belongs on a research or content profile, not on the one that touches servers or deploys files.
Keeping it on one profile also keeps the model settings and any transcription keys in one place.
Members of the AI Profit Boardroom get the installable Agent OS, so the gateway and profiles exist before you add video.
The toolsets reference marks the video toolset as opt-in, so a fresh session will not include video_analyze.
Add it to that profile's config.yaml:
toolsets:
- video
Or start a one-off session with --toolsets video to try it first.
The vision tools source never sends video to the profile's main model; it always goes through the auxiliary vision client, with Gemini named as an example of a video-capable model.
The model is read from auxiliary.video.model first and auxiliary.vision.model second, while the provider comes from the vision block in the configuration docs.
A request to give video its own full backend is open as issue 27313.
If no auxiliary vision client resolves, the tool never registers, so an empty tool list almost always means this step is unfinished.
Supported formats are mp4, webm, mov, avi, mkv and mpeg; files over 20 MB may be slow, and the cap is roughly 50 MB.
For anything already on YouTube, the lighter route is the bundled skill, which the youtube-content docs describe as YouTube transcripts to summaries, threads and blogs.
It works by running a small helper through the terminal tool, so prepare its youtube extra once in the Hermes environment, as the skill's setup section shows.
After that, any profile can call it as a slash command: /youtube-content, then the format you want and the link.
Formats include chapters, a summary, chapter summaries, a thread, a blog post and timestamped quotes.
Because it reads captions only, it needs no vision model, and it cannot see anything shown purely on screen.
🔥 Want the video profile pre-wired with the rest of your stack?
Members get video lessons for each setup, the Agent OS installer with Hermes bundled, four live calls a week and a 3,400+ member community sharing the profiles they run.
The plugin arrived in the catalog through the 15 November 2026 canary, and its catalog entry lists it as community tier, maintained by LuuMoss, version 0.2.2 and verified on Windows only.
hermes plugins enable video-xray, then quit and reopen Desktop.Ctrl+Shift+R./video and untick "Analyze attached videos on send" before attaching anything real.The plugin uses the chat model the Desktop session is already running, adds the video_xray and video_xray_jobs tools, and exports a Markdown report, SRT and JSON cues and a shot-split prompt.
| Route | Input | Output | Needs |
|---|---|---|---|
video_analyze | File path or URL plus a question | A written answer with description | Video toolset, auxiliary video model |
youtube-content | YouTube link or video ID | Summary, chapters, quotes, thread or blog post | The prepared youtube extra |
| video-xray | Attached file and a time window | Report, subtitles, frame gallery, prompt | Windows Desktop, ffmpeg, vision chat model |
Clips posted in an observed Telegram group fit the first route: the Telegram docs say the file is cached and opened with video_analyze only when someone asks, as long as the toolset is on.
If you want your agents to make clips instead of reading them, that is the separate generation side in the Hermes Agent AI video generation guide.
Once the steps above are done, paste these into the video profile.
List your tools and confirm whether video_analyze is available.
If it is, run it on ~/Videos/test-clip.mp4 and tell me what happens
in the first ten seconds and whether any text appears on screen.
Then confirm the youtube-content skill is installed. Change nothing.
/youtube-content Give me chapters with timestamps, a short summary
and the five most quotable lines from https://youtu.be/VIDEO_ID
Save the result as webinar-notes.md in my notes folder.
Use video_xray on the recording I just attached. Analyse minutes 10
to 25 only, 16 frames per segment, transcription engine auto.
When video_xray_jobs shows it finished, list the scene changes with
timestamps and tell me where the SRT and report were saved.
Each prompt reports before it files anything, so you can check the answer first.
Prompt A is the one to run first, because it proves both the toolset and the auxiliary model are working before you rely on them.
Prompt B needs no vision model at all, so it works even on a profile whose main model is text-only.
Prompt C only works on the Windows Desktop machine with the plugin loaded, and the time window keeps each job small.
📞 Want a plan for which videos your agents should read?
Book a free SEO strategy session and my team will map the profiles, the footage and where the reports should land.
video_analyze.Write the routine into the profile's notes, along with which footage may go to a cloud model and which must stay local.
That keeps the setup predictable when a second person starts using the same Agent OS.
It also gives you a simple record of which route handles which job, which helps when you add more profiles later.
video_analyze URLs, and private addresses are refused while allow_private_urls is false.Do Steps 2 to 4 this week on one content profile: they cover quick clip questions and every YouTube link.
Add video-xray only on a Windows Desktop machine, with auto-analysis off, once long recordings become regular work.
Run Prompt A first, and only move on once it confirms both the tool and the skill are in place.
Add video to the toolsets list of one profile and set a video-capable auxiliary model under auxiliary.vision, then start a fresh session.
Either the profile lacks the video toolset, or no auxiliary vision client resolves. The tool only registers when one does.
Yes. The bundled youtube-content skill reads the transcript, so any chat model can turn it into a summary, chapters or quotes.
Only if you run Hermes Desktop on Windows and want subtitles, timestamped reports and exports from long local recordings.
By default, yes. Untick the Analyze attached videos on send option and jobs start only when you ask.
The limit is size, not length: about 50 MB, with a slowness warning above 20 MB. Trim longer recordings first.
Finish Step 3 today, run Prompt A, and book a free SEO strategy session or bring it to the AI Profit Boardroom as you build out your hermes agent video analysis.
About Julian
I'm Julian Goldie, an AI entrepreneur, SEO expert and the founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
My YouTube channel has 400,000+ subscribers, and I run Goldie Agency, a seven-figure SEO agency.
→ The Best Hermes Agent Plugins: My Top 10, And Where Each One Fits In Your Agent OS
→ Hermes Agent Browser: What It Is, The Quick-Start And Where It Fits In An Agent OS
→ Hermes Delegation: Run Parallel Subagents In Your Agent OS With 5 Copy-Paste Prompts
→ Hermes Agent Kanban Orchestrator: Run It In Your Agent OS
The installable system behind every guide on this site — dashboard, agents, memory and pipelines — updated as new tools land, with 3,400+ members and four coaching calls a week.
Join the AI Profit Boardroom →📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉