Hey, I watched your video a few times and really like the idea. Is the inferencing being done on the CPU, do you support GPU as well?
The idea is solid and I like the direction you’re going with it, but the demo doesn’t really show it off. There’s a lot of jumping around in the UI and it’s hard to follow what’s happening without any audio. The interesting bit is right at the end when the rule gets generated, but it’s over so fast that you don’t really get a feel for what Syd is actually doing under the hood.
It was a bit hard to follow with no audio, just a simple “here’s the scan running, here’s the parser kicking in, here’s where the model steps in” kind of thing. Even speeding up the slower parts would make it easier to see the flow. Right now it feels more like a screen recording than a walkthrough. When you’ve spent hundreds of hours inside something it all feels obvious, but for someone seeing it for 3 minutes it’s tough to piece together what’s happening. Been there myself.
The automation angle you mentioned in the post is the part that really sells it. If the tool can take a directory, scan it, parse, correlate and then spit out the rule with almost no manual copying, that’s the kind of workflow improvement I (and maybe others?) care about. The video doesn’t quite show that yet, so it’s hard to judge how smooth the actual experience is.
I’m not against backing something like this, especially as it runs locally and handles the annoying parts. £250 is fine, but at the moment the payment page is just a Stripe form with no real signal that the thing is ready or actively maintained. A clearer demo, a roadmap, or even a short narrated “here’s the state of it today” would go a long way in building confidence.
Apologies if this comes across a bit direct. The idea is solid though. Local LLM + structured output from real security tools is genuinely useful. Keep going.
Really appreciate the detailed feedback—this is exactly what I need to hear.
GPU/CPU question: Yes, Syd supports both. It auto-detects CUDA if available and falls back to CPU if not. With GPU (tested on RTX 3060), inference runs at 30-50 tokens/sec. On CPU it drops to 5-10 tokens/sec, which is usable but noticeably slower for larger responses. The model is quantized (Q4_K_M) to keep VRAM requirements reasonable(6GB).
On the video: You're absolutely right Ive been staring at this for months and forgot what it looks like to someone seeing it fresh. The lack of audio and the jumpy editing makes it hard to follow the actual workflow there are more videos on the website 5 in total I'll
redo the demo with:
- Narration or at least on-screen captions explaining each step
- Slower pacing on the important bits (the parsing LLM rule generation flow)
- A clear "here's the input here's what Syd does here's the output" structure
- Maybe a side-by-side showing manual workflow vs. Syd's automation
The automation is the whole point—scan directory, hit YARA match, auto-parse, explain in plain English, suggest next steps—and the current video completely fails to demonstrate that smoothly.
On the payment page: Fair point. It's bare-bones right now because I've been heads-down on the tool itself, but that doesn't inspire confidence if you're considering backing it. I'll add:
- Current development status (what's working today vs. what's planned)
- Roadmap with realistic timelines
- Maybe a shorter "state of the project" video or changelog
- Clearer communication on what backers get and when you will recieve weekly or monthly updates and obviously ill answer any questions
Current state for transparency:
- Core features working: YARA, Nmap, Volatility, Metasploit, PCAP analysis with RAG-enhanced explanations
- 356k chunk knowledge base indexed and searchable
- Exploit/CVE database integrated
- GUI and CLI both functional
- Still refining: UX polish, additional tool integrations, documentation
I'm actively developing this (clearly evidenced by me responding to HN feedback at [current time 10:38am). The idea of local LLM + security tool orchestration is genuinely useful—I use it daily—but I need to do a better job showing how it's useful and building confidence that it's not vaporware.
Thanks for being direct. This kind of feedback makes the product better. I'll update the demo and payment page this week and can ping you when it's improved if you're interested. and if you sign up on the website thats a great way for me to keep in touch
The idea is solid and I like the direction you’re going with it, but the demo doesn’t really show it off. There’s a lot of jumping around in the UI and it’s hard to follow what’s happening without any audio. The interesting bit is right at the end when the rule gets generated, but it’s over so fast that you don’t really get a feel for what Syd is actually doing under the hood.
It was a bit hard to follow with no audio, just a simple “here’s the scan running, here’s the parser kicking in, here’s where the model steps in” kind of thing. Even speeding up the slower parts would make it easier to see the flow. Right now it feels more like a screen recording than a walkthrough. When you’ve spent hundreds of hours inside something it all feels obvious, but for someone seeing it for 3 minutes it’s tough to piece together what’s happening. Been there myself.
The automation angle you mentioned in the post is the part that really sells it. If the tool can take a directory, scan it, parse, correlate and then spit out the rule with almost no manual copying, that’s the kind of workflow improvement I (and maybe others?) care about. The video doesn’t quite show that yet, so it’s hard to judge how smooth the actual experience is.
I’m not against backing something like this, especially as it runs locally and handles the annoying parts. £250 is fine, but at the moment the payment page is just a Stripe form with no real signal that the thing is ready or actively maintained. A clearer demo, a roadmap, or even a short narrated “here’s the state of it today” would go a long way in building confidence.
Apologies if this comes across a bit direct. The idea is solid though. Local LLM + structured output from real security tools is genuinely useful. Keep going.