Build your research portfolio

It turns a question into a research thread that can stay open after one run ends. The harness searches the public web, keeps the sources it selects, reads each selected source, and saves a separate research artifact for every source that it can successfully process.

A completed source artifact includes the source title and link, a grounded summary, why the source matters to your question, key claims, supporting evidence, recommendations, and limitations. A source that cannot be loaded or extracted is recorded as incomplete instead of being presented as a finished result.

No. A thread keeps its runs, selected sources, and completed artifacts as durable records. You can add another prompt to the thread or continue eligible unfinished work later. Earlier artifacts are not silently copied into every new prompt; you can explicitly choose a bounded set of artifacts when a later run needs them.

You can ask about any topic with useful public material, but the result depends on what the search can find and what the retrieved pages actually support. The harness reports what the run covered and what gaps remain; it does not promise that one run will find every relevant source or settle a question by itself.

Your research thread is tied to your Vidbyte account and can only be read through an authenticated browser session or API request for that account. It is not published as a public resource. Questions about how outside providers retain or use data belong in the applicable privacy and provider terms.

The harness runs several focused web searches. It keeps the relevant results, avoids repeating links it has already used for the thread, and continues toward the run's source target while the approved search and capacity limits remain available.

It reads public web pages that the search can reach. A page must be publicly available and contain usable text. Login walls, unavailable pages, empty pages, and pages that are too large can be left out instead of being treated as readable evidence.

The source is set aside while the harness reads its page. If the page has usable text and is not too large, a separate reading step turns it into a structured result. Only a result that passes that check is saved as a completed source in your thread.

It reads one source at a time and does not fill gaps by guessing or browsing elsewhere. The result separates the source summary, why it matters, claims, evidence, recommendations, and limitations. The run also records what it covered and what remains open, so one source is not presented as proof of the whole answer.

Yes. The harness saves progress as it finds and reads sources. A later prompt adds to the same thread, while a continue action resumes unfinished work. Finished sources are skipped, and sources that could not be completed can be tried again when appropriate.

A source-level problem does not turn that source into a completed artifact. The run records a bounded failure reason, continues independent work where it can, and returns partial progress when the run cannot finish. The saved work remains available for a later prompt or continuation instead of being discarded.

The signed-in browser surface uses the browser-session research routes. A script or integration using the public research API must authenticate with a Vidbyte API key. Public API requests use the API-key identity to scope the thread and its reads to the owning account.

The public API can create a thread and run, append a prompt to a thread, continue a run, poll run status, list the account's portfolio, and read one owned thread. The mutation routes return an accepted run identity; execution happens durably after admission, so clients should poll status rather than expect the final artifacts in the create response.

The run uses the account's API wallet. Vidbyte authorizes capacity before billable model, search, and fetch work, records the SDK usage and provider operation attempts in the run's usage session, and settles cumulative usage at durable progress and terminal boundaries. A page served from the shared content cache does not create a new content-provider call.

The harness stops before starting work that the wallet or approved run budget cannot cover. Completed artifacts and durable progress remain available, and the run reports a partial or capped outcome. After the account has capacity, the public continue operation can resume eligible unfinished work instead of starting the completed sources again.

One Research Summary is one completed structured artifact for one source. It is counted by finished source work, not by the number of words generated. The artifact carries the source identity, findings, relevance, recommendations, evidence, limitations, and completion state.

Sources differ in length, accessibility, and how much model and provider work they require. The page therefore presents capacity as an estimate and shows the requested scope and cap before an expensive run begins. The actual result can be partial when sources fail or the approved capacity is reached.

The configured run presets target 15 sources for small, 50 for medium, and 100 for large. A request can provide its own target within the application ceiling, which is currently 1,000 sources. The target is a ceiling for the run's discovery and extraction work, not a promise that every source will produce a completed artifact.

The run pauses before the next operation that would exceed its approved budget, saves the progress already completed, and reports the completed count and remaining work. Adding capacity or continuing the run uses the durable checkpoint rather than throwing away the artifacts that already finished.

No. The price cap is checked before paid work starts, and the run pauses when it reaches that cap. Retries and additional reading work count against the same approved amount rather than being silently added afterward.

The harness can reuse a cleaned-up version of a recently fetched page for up to 60 days, so follow-up work may not need to fetch it again. The finished research summary stays in your thread, while the original HTML and PDF are not stored as part of that summary.