Hey everyone, Moe here.

We're building Webhound — AI research agents where you set a budget and the agent keeps researching until it runs out. More budget = more sources, more depth.

How we got here: We started with a product (now called Datasets) that builds structured tables from the web. It worked, but we hit a wall: as datasets got bigger, costs exploded and context became unmanageable. And for open-ended datasets, we never had a clean answer for "when should the agent stop?"

So we rebuilt everything around two ideas: costs that stay linear with time, and giving users direct control over depth.

The new architecture is a Planner-Executor-Verifier loop. Executors get fresh context each run. Planner and Verifier work from summaries and can search over what's accumulated. Runs until budget's out.

We tested it on traditional deep research as a sanity check. At $3-5, it beat the top result on DeepResearch Bench. At $15, it way outperformed it, even running on older models (gemini-2.5-flash). So we shipped it as Reports and put the same architecture into Datasets.

We now have two products:

• Reports = cited research docs (competitive analysis, market research, system design). Always uses the full budget—more budget means more sources and deeper coverage.
• Datasets = build structured tables from the web, or upload your own and enrich with new columns. We can access LinkedIn, YouTube, Reddit, and most other sites. Examples: lead lists with company info, competitor feature matrices, lists of VCs filtered by thesis, GTM leads with emails. Stops when complete, or at budget, whichever comes first.

Asks:

• Try it – webhound.ai, $5 free. Honest feedback welcome.
• Spread the word – if it's useful, share it with anyone you know doing a lot of research (investors, analysts, consultants)

– Moe

(Demo Video for Reports: https://www.youtube.com/watch?v=eEGJBudvmsE)