Browse categories
Explore
Fiverr Pro
English
$
USD
I will lock down what your ai coding agent can access on your machine
Tyler L
About this gig
Your AI coding agent can read your files, run shell commands, and reach the network - usually without you ever seeing what it actually touches. That's useful, and it's a real attack surface: a bad prompt injection or a compromised tool call can leak data or do damage, and most setups have zero visibility into it.
I set up real, working controls - not a checklist, an actual configured system:
Interactive outbound monitoring (OpenSnitch or equivalent) scoped to your actual agent tools, catching genuinely new connections without nagging you on every legitimate call
A reasoned trust policy: which processes get standing access, which stay on a short leash, and why. Blanket-trusting a shared interpreter like python defeats the whole point - I don't do that
Zero-trust remote access setup (Twingate or equivalent) if you want to safely reach services on your network without opening ports
A written report on what each agent can actually reach and how to extend the policy yourself
Every rule is built and tested against your actual tools, live, on your machine.
Get to know Tyler L
Tyler L
Freelance AI Pipeline Engineer
- FromUnited States
- Member sinceJun 2020
- Avg. response time1 hour
Languages
English
I am a freelance AI pipeline engineer specializing in local generative-AI systems, including ComfyUI workflow development and image-to-3D asset production. I build and repair pipelines that run reliably outside of cloud infrastructure, focusing on version-pinned and VRAM-aware solutions. With a background in technical troubleshooting and corporate training, I deliver systems complete with documentation and client instruction.
When you choose me, you choose a guy who values his clients, and you'll get a personal one on one connection and feel like your problems matter.
Other AI Development Services I Offer
FAQ
Will this block my agent from doing its normal work?
No - that's the actual point. The goal is safe, not neutered. Standing rules get built for your real tools so they keep working uninterrupted; only genuinely new or unexpected activity gets flagged.
What if I'm on Windows or macOS, not Linux?
Standard and Premium still apply - the tools differ (Little Snitch on macOS, comparable options on Windows) but the same reasoned-policy approach applies. Flag your OS up front so I scope the right tooling.
Do you need access to my actual machine?
Premium includes a screenshare session for the live setup. Basic and Standard can often be done from your exported config plus a written runbook you apply yourself - ask if you'd prefer that.
