# Webhound — Budget-capped autonomous web research returning cited reports, datasets and machine-readable claims via MCP or API. > Source: The Agents Index — https://theagentsindex.com/webhound (structured, researched, re-verified) > Facts last verified: 2026-09-27 Webhound is an autonomous web-research engine that takes a question and a dollar budget and returns cited reports, sourced datasets, and JSON claims carrying their evidence. It is driven from an agent over MCP or a REST API, or from its own hosted workspace UI. | Fact | Value | | --- | --- | | Website | https://www.webhound.ai | | Price | from $1/per research run (minimum $1 budget) pay-as-you-go | | Free trial | Partly | New accounts receive a one-off $5 grant — 'New accounts include one $5 Report or Dataset' (pricing page); the homepage states '$5 free on signup — about 75 minutes of research'. No standing $0 plan; after the grant all use is prepaid credit. ## What it does **Writes code** · Partly — It writes and runs Python against your own research data for analysis and charts, but it is a research engine, not a tool that authors application code from a spec. > Run code. The agent writes and executes Python for data analysis, charts, transformations, and file generation—using your actual datasets, not made-up data. > — https://www.webhound.ai/news/webhound-agent · Webhound Agent announcement, 'What it can do' list, 'Run code' bullet **Done-for-you service** · No — Self-serve software you run yourself from an agent or the web app; nothing on the site offers people who run research engagements for you. > Webhound is your agent’s research engine — made to be called by an agent and left to run. When you want to get hands-on, the same engine runs behind a rich interactive UI. > — https://www.webhound.ai/ · Homepage, 'What it is' section introduction ## How much it does unattended **Runs autonomously** · Yes — The stated default is an unattended run: an agent starts research by tool call, the engine keeps working to the budget, and results return as JSON. > An agent starts research on its own · Wire it into a pipeline or a tool call · Results come back as structured JSON · No human has to sit in the loop > — https://www.webhound.ai/ · Homepage, 'Your agent drives — MCP & API, the default' card bullets **Multi-agent** · Partly — Chains run several research and extraction steps in sequence, each consuming the previous step's output, which is orchestration of stages rather than several agents coordinating concurrently on one task. > Describe a sequence of steps — "research the best CRMs, then extract 30 of them with pricing, then write a buying recommendation" — and the system runs them automatically in order. Each step builds on the prior step's output. > — https://www.webhound.ai/news/webhound-agent · Webhound Agent announcement, 'Multi-step pipelines' section **Agent permissions** · Partly — The dollar budget is a hard ceiling set before the run and pipelines need plan approval, but the vendor documents no per-capability permission controls over what the agent may touch. > You approve the full plan before anything starts, and you can stop the pipeline at any point. > — https://www.webhound.ai/news/webhound-agent · Webhound Agent announcement, 'Multi-step pipelines' section ## Which models it runs on **Claude** · No — The Hound harness is stated as built with DeepSeek V4 Pro and GPT-5.4 and is not a selectable model; Claude appears only as an MCP client that calls Webhound, not as an inference option. > Hound 1.0 is Webhound’s research harness, built with DeepSeek V4 Pro and GPT-5.4. Set a dollar budget and it spends that effort on the investigation. No subscription, no seats. > — https://www.webhound.ai/ · Homepage, 'Hound 1.0 — Pay only for what you use' section **GPT** · Yes — GPT-5.4 is named by the vendor as one of the two models the Hound 1.0 research harness is built with, though buyers cannot select it themselves. > Hound 1.0 is Webhound’s research harness, built with DeepSeek V4 Pro and GPT-5.4. > — https://www.webhound.ai/ · Homepage, 'Hound 1.0 — Pay only for what you use' section **Open models** · No — A DeepSeek model sits inside the harness, but the vendor states Hound is not a selectable model or mode, so a buyer cannot choose or supply open-weight models. > Webhound now exposes Hound, a research harness built with DeepSeek V4 Pro and GPT-5.4 across planning, execution, verification, and assembly. Hound is not a selectable model or mode and does not resolve to a single provider backend. > — https://www.webhound.ai/news/session-minimums · Archived product note banner at the top of the session-minimums post **Model choice** · No — The vendor chooses the models: Hound is explicitly not a selectable model or mode, and the older Flash/Pro model picker was retired, leaving only the dollar budget as a dial. > Hound is not a selectable model or mode and does not resolve to a single provider backend. You choose the question and dollar budget; the budget controls research effort. > — https://www.webhound.ai/news/session-minimums · Archived product note banner at the top of the session-minimums post ## Where you use it **In your editor** · Partly — It reaches editors only as an MCP tool inside a client such as Cursor or Claude Code; the vendor documents no native IDE extension of its own. > Webhound runs as a tool call inside Claude Code, Codex, Manus, and any MCP client. Kick off a research job, keep working, and read structured results back without leaving your agent. > — https://www.webhound.ai/ · Homepage, 'See it run — Inside your agent' section **On the command line** · Partly — There is no Webhound CLI; the terminal surface is borrowed from CLI agents such as Claude Code and Codex that call it over MCP. > Webhound runs as a tool call inside Claude Code, Codex, Manus, and any MCP client. > — https://www.webhound.ai/ · Homepage, 'See it run — Inside your agent' section **In a browser** · Yes — A hosted web workspace is a first-class surface, with a chat agent home page, nested folders, session steering and published results alongside the API. > Read and verify in a rich UI · Steer a run mid-flight · Publish and share results · Explore claim traces by hand > — https://www.webhound.ai/ · Homepage, 'You drive — the UI, when you want control' card bullets _Not established: In your pipeline_ ## Whose machine it runs on **Self-hosted** · No — Everything runs in the vendor's cloud: the documented integration is a hosted REST endpoint with a Webhound API key, and no self-hosted or bring-your-own-infrastructure option appears anywhere on the site. > Full programmatic access to research, extraction, publications, and workspace management — https://api.webhound.ai/api/v2 · 58 endpoints · Bearer auth · JSON responses > — https://www.webhound.ai/docs/api · API v2 docs, page header and base-URL strip _Not established: Open source_ ## What it costs to run **How it meters** · Yes — Purely consumption-priced: you prepay credits and spend a per-run dollar budget, with per-operation and per-token rates published rather than a seat or subscription fee. > Prepaid Credits: Add credits to your account before using Webhound · Real-time Deduction: Costs are deducted as operations are performed · No Subscriptions: Pay only for what you use, with no monthly fees > — https://www.webhound.ai/news/pay-as-you-go · Pay-as-you-go announcement, 'Simple, Transparent Pricing' list **Free tier** · Partly — You can start without talking to sales, but the free path is a one-off $5 grant on signup rather than a standing free tier you can keep using. > New accounts include one $5 Report or Dataset. > — https://www.webhound.ai/pricing · Pricing page, note beneath the 'Example budgets' grid **API access** · Yes — A documented public REST API with bearer keys covers research, extraction, pipelines, files, publications and account credits, and it is presented as the default way to drive the product. > All requests require an API key in the Authorization header: Authorization: Bearer wh_your_api_key_here. Get your API key from the Account page. Keys start with wh_ > — https://www.webhound.ai/docs/api · API v2 docs, 'Authentication' section **MCP server** · Yes — MCP is the vendor's stated default interface, letting any MCP client start a run and read back structured results without a person in the loop. > MCP & API — the default … An agent starts research on its own … Results come back as structured JSON > — https://www.webhound.ai/ · Homepage, 'Your agent drives' card heading and bullets **Bring your own key** · Partly — You save your own third-party API keys and the agent calls them mid-run; inference keys are not a thing here because the models are not selectable. > Yes. Save an API key once and the agent generates and runs the calls at session time — your keys stay private to your workspace. > — https://webhound.ai/ · FAQ, 'Can it use my own data and API keys?' ## Buying it for a team **Seat model** · No — There are no seats to buy at all: the vendor sells prepaid research budget per run, so cost scales with work done rather than with headcount. > Set a dollar budget and it spends that effort on the investigation. No subscription, no seats. > — https://www.webhound.ai/ · Homepage, 'Hound 1.0 — Pay only for what you use' section **Audit log** · Partly — The API exposes a per-session work log and account spending history, which records what the agent and your credits did, but nothing described covers administrative actions by an organisation's users. > GET /sessions/:sessionId/tasks Work log … GET /account/usage Spending history > — https://www.webhound.ai/docs/api · API v2 docs, 'Session Reading' and 'Account' endpoint tables _Not established: A company can buy it, Pooled budget, Admin controls, Single sign-on_ ## What happens to your code **Getting out** · Partly — Outputs leave cleanly in ordinary formats and via an export endpoint, but the vendor states nothing about whether unused prepaid credits expire or can be refunded. > Cited long-form analysis. Every claim links to the page and the tool call behind it. Export to Word or HTML. … You define the columns; rows come back as CSV, JSON, or Excel. > — https://www.webhound.ai/ · Homepage, 'Outputs' section, Reports and Datasets cards **Certifications** · Partly — GDPR and UK GDPR legal bases and data-subject rights are set out in the privacy notice, but no SOC 2, ISO 27001 or other certification is claimed anywhere on the site. > The General Data Protection Regulation (GDPR) and UK GDPR require us to explain the valid legal bases we rely on in order to process your personal information. > — https://www.webhound.ai/privacy · Privacy Notice, section 3 'What legal bases do we rely on', EU/UK paragraph _Not established: Opt out of training, Data residency_ ## Provenance Every fact above comes from the vendor. Nothing here is independently corroborated yet. ## Coverage elsewhere None recorded. Every fact above comes from the vendor and is not independently corroborated yet. ## What changed - 2026-09-27 — Capabilities · Commercial terms · Free tier · Price - 2026-09-03 — Capabilities · Commercial terms · Free tier - 2026-08-31 — Capabilities · Commercial terms · Free tier · Price - 2026-08-23 — Alternatives · Crunchbase link · Documentation link · LinkedIn link · One-liner · Summary ## Correcting this record Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Webhound, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth. Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.