Webhound

Webhound is a research agent meant to be called programmatically, for example as an MCP tool inside Claude Code, Codex, or Manus, or used directly as a standalone app. It treats research depth as a dial rather than a fixed setting: tell it how much a question is worth, and it spends that budget digging until it hits the limit, instead of stopping after a fixed number of searches. Output comes back as either a structured dataset or a fully cited report, with every claim linked back to the source and tool call that produced it, which matters for anyone who needs to show their work rather than just get an answer.

It is pay-as-you-go with no subscription, which suits both individual power users doing occasional deep dives and developers wiring research into agent pipelines where usage is bursty. Compared to the built-in 'deep research' features bundled into general chat assistants, Webhound's pitch is being agent-native infrastructure first and a hosted app second, making it a dependable research primitive for teams building on top of LLM agents rather than a standalone research UI for casual browsing.