I've been seeing a wave of AI "coaching" and "analysis" tools for CS lately, and almost all of them run the same play: slap a chat window on top of some stats, charge 10 to 20 bucks a month, and tell you things like "your utility usage could improve." Most of them give you nothing you couldn't figure out from the scoreboard. And even the ones that do surface something real, honestly, it's not information worth a subscription. That bugged me enough that I built the thing I actually wanted and I'm just giving it away.

Some background so you know where this comes from. I'm a software developer and I've played FACEIT with the same 5 stack for years. We're the kind of team that downloads opponent demos before a league match and then nobody actually watches them, because sitting through three demos at 8x speed taking notes is miserable.

So this tool does the demo homework. You drop in demos of the team you're about to play and it parses every round on your own PC. It works out their CT setups and who anchors where, their utility lineups and the timings they throw them on, eco and force habits, how they rotate after first contact, what falls apart when they lose a couple rounds in a row. You get a scouting report per team and map, plus a chat where you ask things like "what does their mid player do after a lost pistol" and it answers from the parsed rounds with sample sizes attached. If it only saw something twice, it says so. Being able to trust a claim matters more than the claim itself.

One thing it deliberately doesn't do: aim. It won't tell you your crosshair placement is bad. There are a hundred tools for that. This is for playing better as a five and reading the enemy team, which in my experience is where matches at our level are actually won.

A detail I care about a lot: callouts. The default zone names come from the game files, and nobody in a real stack says "CTSideUpper". You can rename every default callout and add your own zones that don't exist in the game at all. Our stack has a "ninja" spot on Anubis B, so I added it, and the analyst just uses it in answers like it's always been there. Set your callouts before you start asking questions, it makes every answer land better.

Fair warning on the AI part, since it would be weird not to mention it here: I used AI end to end while building this. I'm a dev and I reviewed what shipped, but a lot of the code is AI-assisted. It runs locally on Windows, seven maps work completely out of the box (Ancient, Anubis, Cache, Dust2, Inferno, Mirage, Nuke), and the chat uses your own API key. Gemini's free tier covers normal use, so the running cost is zero or close to it. Your demos and data never leave your machine.

Repo is here, free for personal use, and the readme walks you from installing uv to your first scout even if you've never touched Python: https://github.com/Git-Uzair/counterstrat

The reads get scary at three or more demos of the same team on the same map. One demo just gives you hints. If you try it with your stack I'd genuinely like to hear what reads are missing, that feedback is the whole reason I'm posting.

Problem is paid game coaching targeted towards the general playerbase has mostly always been a scam. Used back then by high level players as a way to generate needed income. While informing the player with generic insight. :Insert Prop Joe quote about relationship counselors:

When heat maps were introduced around 2018. It was a quick way to inform someone of players positional tendencies. Since then its morphed into including other basic metrics. To the point its making players think that consuming spreadsheets will make them better players. Factor in AI, its just repackaging other services who are using the same generic data points.

Realistically such tools are only beneficial for low tiered teams participating in leagues. For their problem is the IGL's lack of time. While at the Pro level its relegated to a dataset collection generator for support coaching staff to find validity in.

Overall playerbase who use these services are wasting their time.

I've actually used some new research papers on representing 3d and time series data as text for llms which might give better results. And also you're spot on about paying for these services.

Not trying to cast disparity directly at you. In my 10+ years of posting here and consuming CS content. Ive yet to see someone properly break down how to aim train correctly.

So i have no faith in your LLM being different.

This is tool is not for aim training, but I agree, this is far from optimal but it works for me, you can give it a shot and see for yourself.

Hi I'm having trouble with the first read. It says first read failed; map card for de_ancient not found. I'm using Gemini 3.1 flash.
Really interested in the project myself. Any more features you're looking to add?

I'll fix this. By tonight, you can suggest more features

Thanks! I suggest a viewer to check the evidence. Perhaps something similar to cs2d.app

Can you try with other maps, is it working?

Unfortunately no. Nuke isn't working as well

Debugging and fixing right now, I'll comment here

It's working for my friend tho

Cool will try this, thanks!