The problem
Agents are trained to complete tasks. That works well when completion is observable: the code runs, the test passes, or the file exists. Research is different. There is rarely one natural stopping point. The same question can have a ten-minute answer, a two-hour answer, or a twenty-hour investigation, and each can look complete on the surface.
Most agents therefore stop at the earliest point where they have enough information to answer confidently. But confidence is not completeness. For questions that affect a company, a career, an investment, an investigation, or another important decision, the missing work is often where the value lives.
Our belief
The missing primitive in agentic research is explicit control over effort.
Users should be able to answer a second question alongside the prompt: How much work do I want the agent to do? Webhound makes that effort legible by mapping budget to research time.
Budget is not merely pricing. It is a depth dial.
What Webhound is
Webhound is a research engine for questions where you want to control how much work goes into the answer. It starts with two inputs: a prompt and an effort budget. The budget acts as fuel for a broad, multi-source investigation, which Webhound assembles into a usable output.
Webhound is built to be used by your agent first. Your agent can hand Webhound a research task, choose how much effort it deserves, and use the result without having to supervise every search or repeatedly ask it to go deeper.
The final answer is not the entire product. A Webhound run also produces working documents that show the research taking shape. Your agent—or you—can inspect and use that work instead of receiving only a black-box answer.
Who it is for
Webhound is defined less by a traditional ICP than by the value of the question. It is for anyone facing a question where missing information costs more than the research: choosing a company or market, finding manufacturers, evaluating an executive, understanding a policy problem, uncovering journalistic leads, mapping a technical design space, or making a consequential personal decision.
Not every question needs Webhound. Simple questions should use simple tools. Webhound matters when controlling the amount of research can change the result.
Our commitments
- Depth must be controllable. The user—not the agent's default stopping behavior—decides how much effort the question deserves.
- The work must be inspectable. Working documents, sources, and the research trail are first-class product surfaces.
- More work must create more value. We optimize for better conclusions and discoveries, not activity, verbosity, or the appearance of rigor.
- The economics must stay honest. No subscription designed to profit from inactivity. Webhound wins when people find the research useful enough to run again.
- Breadth must still feel concrete. We can serve many kinds of users, but every explanation and product experience should make an immediate real-world use obvious.