How I Used TinyFish to Find Local Suppliers I’d Have Missed
About the builder
Gia-Thi Nguyen is a digital transformation leader with decades of experience across Siemens and SAP, including roles such as CIO, CFO, and Head of Solution Advisory. He helps complex organizations realize value by strengthening their capabilities across data, applications, processes, and people.
TL;DR
I used TinyFish with my personal AI assistant, Molthi, to research local material suppliers in Vietnam for a personal project.
The first run required 1 search and 1 fetch at $0 cost.
But the more meaningful result was not simply time saved. TinyFish helped turn a fragmented, language-heavy sourcing task that I realistically would not have completed manually into something I could actually act on.
“This was not just about making an existing workflow faster. It was about making a task possible that I realistically would not have completed manually.”
The problem
Local sourcing sounds simple until you actually try to do it.
For a personal project in Vietnam, I needed to identify material suppliers in a regional market.
Not theoretically.
Practically.
Who sells the materials I need? Which suppliers are relevant enough to investigate further? Which sources look credible enough to call, visit, or add to a shortlist?
The problem was not that there were no suppliers.
The problem was that the information was scattered in exactly the way local information often is.
Some suppliers had websites. Some had social pages. Some appeared in local directories. Some had Vietnamese-only pages. Some surfaced through older listings or thin search results that looked unhelpful until inspected more closely.
I speak Vietnamese, but I do not read it fluently enough to scan large volumes of local information quickly.
And that matters.
Search is not just typing words into a search bar. It is deciding which fragments are worth trusting, which pages are dead ends, and which small signals deserve a second look.
Doing that manually would have been tedious.
More honestly, I probably would not have completed it properly at all.
What I built
I use a personal AI assistant called Molthi, built on OpenClaw.
Molthi already understands the broader context of the projects I am working on. I interact with it through WhatsApp, so I wanted to see whether TinyFish could give it a useful web-research layer without turning the task into a technical project of its own.
The instruction was deliberately simple:
“Research TinyFish, here is the API, and I need you to find local sources in Vietnam for the materials I am looking for.”
That was essentially it.
No elaborate architecture. No unnecessary complexity. Just one practical sourcing question and a web environment that was difficult to navigate manually.
The workflow looked like this:
- I gave Molthi the sourcing goal and project context through WhatsApp.
- Molthi used TinyFish for search and retrieval.
- TinyFish helped surface relevant local sources from fragmented web results.
- Molthi organized the findings into a single-page sourcing view.
- I used that page as a decision board for what to investigate next.
The initial result was almost offensively simple:
1 search. 1 fetch. $0 cost.
But the output was useful because it moved the task from scattered research into an organized view I could act on.
Where TinyFish fit
TinyFish fit into one specific part of the workflow: the search and retrieval layer.
This was not a situation where I wanted to build a scraper, maintain brittle selectors, or design a workflow that was more impressive than the problem deserved.
The real question was much simpler:
Can an AI assistant help me find useful local suppliers in an information environment that is messy, fragmented, and partly outside my reading comfort zone?
TinyFish’s job was to:
- search across fragmented local sources
- retrieve relevant web information
- help surface supplier candidates that were not obvious from a single manual search
- make the local web easier for Molthi to process
Molthi’s job was to:
- understand the project context
- organize the research results
- turn scattered findings into a single-page sourcing view
- present the results in a format I could actually use
My job was still to make the final sourcing decisions.
TinyFish did not decide which supplier to use.
Molthi did not replace local judgment.
The workflow simply made it possible to reach a reasonable candidate set in the first place.
The difficult work was not evaluating a supplier after it had been found. I could do that.
The hard part was reaching a reasonable candidate set in the first place.
TinyFish moved the task from:
“I should probably do this someday.”
to:
“I can actually start now.”
What worked
1. Keeping the workflow simple
This experiment did not begin with a large technical build.
It began with:
- one practical problem
- one assistant that already understood the context
- one API that could handle the difficult research layer
- one webpage that made the output easier to use
That simplicity was part of why it worked.
A lot of AI workflows become unnecessarily complicated because people begin by designing the system instead of respecting the task.
Here, the task was clear enough:
Find plausible local suppliers and organize the results into something actionable.
2. Using TinyFish as the research layer, not the final decision-maker
TinyFish helped surface relevant local sources.
Molthi organized the findings.
I still retained the final judgment.
That division of labor made sense because sourcing decisions are contextual. The system did not need to replace human review. It needed to make human review possible without forcing me to manually reconstruct the local web from scratch.
3. Designing the output around the next decision
The raw web is not a good interface for decision-making.
A long list of links is still work.
I needed the findings to be organized in a way that made the next decision easier. Molthi incorporated the research results into a single-page web view that grouped sourcing options into a practical shortlist.
Instead of forcing me to reopen tabs and reconstruct the logic later, the page summarized materials and supplier directions with clear labels such as:
- Promote
- Keep
- Confirm
- Use carefully
The goal was not to automate the final sourcing decision.
The goal was to move from scattered information to a structured starting point.
Measurable outcome
This was not a large benchmark.
It was a lightweight personal workflow that showed how much leverage a small amount of web research can create when the output is organized well.
| Metric | Result |
|---|---|
| Initial TinyFish usage | 1 search, 1 fetch |
| Initial TinyFish cost | $0 |
| Output format | Single-page sourcing view |
| Source environment | Fragmented local web sources in Vietnam |
| Final decision-maker | Human review |
| Identifying details | Removed or anonymized for publication |
The most important result was not the number of minutes saved.
The important result was that a task I probably would have avoided became doable.
People often talk about automation in terms of time saved.
That is not wrong.
But in this case, it is too small a framing.
I could say TinyFish saved me hours of browsing. It did.
But the more honest answer is that without it, I probably would not have done the research properly at all.
Maybe I would have searched for a while. Maybe I would have found two or three obvious suppliers. Maybe I would have stopped once the work became repetitive, ambiguous, and locally messy.
That is the real threshold.
“TinyFish did not just optimize the research. It lowered the friction enough that the task became worth attempting.”
The interesting impact of tools like TinyFish is not always optimization.
Sometimes, the value is that they change what people are willing to attempt.
They make tasks move from ignored to doable.
That is a much bigger category than productivity.
What I’d do differently / Lessons learned
1. Local information can be abundant and still be difficult to access
The web contains plenty of useful local information.
But it is often not packaged for clean discovery.
Small businesses may have exactly what you need while still being almost invisible to normal search habits.
They may not rank well. They may not write in English. They may not maintain polished websites. They may exist as fragments across platforms.
A useful web-research layer helps recover those fragments and bring them back into consideration.
2. Valuable AI workflows do not need to be complicated
This experiment did not require a grand architecture.
It required one real problem, one assistant with context, and one reliable search and retrieval layer.
That was enough to create value.
3. Automation expands the set of problems people bother solving
Saving time is useful.
But the bigger shift happens when a tool changes the effort calculation.
A task that once felt too fragmented, too repetitive, or too locally opaque suddenly becomes worth attempting.
Once that happens, the downstream effects are larger than the original search.
You do not just complete a task faster.
You start taking on tasks you previously avoided.
Recommendation for other builders
Start with a research task you have been avoiding because the information is too fragmented, too local, or too annoying to collect manually.
Do not begin by designing the most sophisticated possible agent workflow.
Begin with one practical question.
Give your assistant enough context to understand the goal.
Then use TinyFish for the search and retrieval layer.
For me, the question was simple:
Can I find useful local suppliers without spending days manually digging through the web?
The answer was yes.
And once that worked, the more interesting question became obvious:
What other tasks have I written off, not because they are impossible, but because the search cost was too high?
Try it / Links
Have a research task you keep avoiding because the web is too fragmented?
Start with one practical question, give your assistant enough context, and use TinyFish for the search and retrieval layer.
📌 Sign up free → agent.tinyfish.chat
Docs → docs.tinyfish.chat
Open source Cookbook → github.com/tinyfish-io/tinyfish-cookbook
