Browse categories
Explore
Fiverr Pro
English
$
USD
I will build an ai agent with tool calling, memory and production deployment
Logan
About this gig
I build the system around the agent - the tools it can call, the memory it keeps, the step where a human approves, and the panel that shows what it cost. An agent without those is a demo, not something you can run on Monday.
WHAT I HAVE ACTUALLY SHIPPED
- Won the AX AI Hackathon 2026 in South Korea by turning a 5-hour manual job
into 3 minutes
- A pipeline I still operate: 56 API routes, 14 tables, 9 per-field rule sets
- Cut per-item AI cost in half by batching - 32 items for $2.08, measured
- A scheduled collector pulling from 36 countries, running today
RULES I DO NOT BREAK
- The agent stops before anything irreversible. A person presses send
- Every external call has a timeout and a fallback
- You see cost per run, not a surprise invoice at month end
- Model choice and its price are written in the plan before I build
HOW IT WORKS - no long back-and-forth
1. You describe the job step by step in the requirements form. I give you the
template, you fill the blanks
2. I return a written plan: what the agent does, what stays human, what it
will cost per run
3. You approve. I build, deploy and hand over the source
I work asynchronously and in writing. No calls required.
Get to know Logan
Logan
Full Stack Web and AI Development, Plan to Deploy
- FromSouth Korea
- Member sinceJul 2016
Languages
English, Korean
I built and run CCgather, the global Claude Code usage leaderboard: 186 developers across 27 countries, 2.64 trillion tokens of usage tracked to date. I rank #3 on it worldwide, so I push these tools as hard as anyone I build for. I won the AX AI Hackathon 2026 in South Korea by turning a 5-hour manual task into 3 minutes, and I build that same automation for clients: AI pipelines that remove repetitive work, plus the product around them. Planning, design, development, deployment. Next.js, TypeScript, Supabase, Cloudflare.
Other Software Development Services I Offer
FAQ
How is this different from a chatbot?
A chatbot answers. An agent acts - it queries your systems, produces the document, files the record, and moves to the next step. The chat window is optional.
Which AI models do you use?
Whichever fits the job. I name the model and the expected cost per run in a written plan, and you approve that before I build anything.
Who pays for the model API usage?
You do, on your own account. I never bill model usage through me. Share a key or I will walk you through creating one.
What counts as one tool?
One action the agent can call, such as write a row to a sheet, create a ticket, or check stock. Read and write count separately.
Can it run unattended?
Yes, on a schedule. If a run fails you get an alert. Anything irreversible stops and waits for a person, so an unattended run cannot send or charge on its own.
What if the agent gets something wrong?
That is what the approval gate is for. From Standard up, a person reviews before anything irreversible goes out, and the decision is logged.
What do you build it on?
Cloudflare Workers, D1 and R2. The agent runs on the edge with scheduled triggers and object storage built in, so there is no server to rent and no monthly infrastructure bill handed over to you.
How do you keep running costs down?
Batching instead of realtime calls. On my own pipeline that brought 32 items down to $2.08, measured. Premium adds a budget cap.
Do I get the source code?
All of it, plus a run guide. Nothing is tied to a server of mine.
Do we need a call?
No. Requirements form, then a written plan for your approval, then delivery. Everything in writing, on your schedule.
