Clear answers. Before your first run.

Start with a single question in your workspace. Choose an effort level, add permitted context if needed, then inspect the report and its evidence. Use a ready-made agent or leave routing on Auto.

Use a saved agent for a reusable method, batches for many input rows, schedules for recurring checks, or a playbook for a coordinated workflow. Availability and limits depend on your workspace plan.

Yes. Add workspace files or import a supported URL and attach the relevant context to a run. Check supported formats, extraction quality and current limits before relying on a document-heavy workflow.

Proofline picks the model for every step itself: a fast one for quick lookups, a balanced one for most research and the strongest for the deepest verification. You only choose the effort.

Yes. Verification helps catch unsupported statements, but it is not a guarantee. Source excerpts and reviewer corrections keep the evidence inspectable.

The system reopens cited sources and looks for the quoted passage. A passage match helps inspect a claim; it does not establish that every inference or real-world conclusion is correct. Review disputed and incomplete findings.

A stopped job can keep a partial answer. Inspect status and stop_reason, review what was collected, and decide whether a follow-up or supported continuation is appropriate. Do not present a partial answer as a completed check.

No. Report completion and an external write are separate operations. Review proposed actions and the approval rules configured in your workspace; some policies can allow selected actions automatically.

The current report export API supports Markdown and JSON. An export can be processing before it is ready. PDF, Word and slide exports are not offered by the current export contract.

Your workspace controls its sources, access and retention. Read the Trust Center for deployment and data-handling details.

Workspace roles separate administration, building, reviewing and reading. API integrations use workspace-bound keys with explicit scopes. Review the creator’s role, key mode, required scope and plan availability together.

No SOC 2 Type I, SOC 2 Type II or ISO 27001 certification is claimed. The Trust Center publishes the current status and the questions to resolve before a production deployment.

Workspace administrators can configure retention for files, audit history, runs and reports. Available choices differ by data type. Review the Trust Center and deployment requirements before setting a policy.

Each effort has a published most per job, shown before it starts. You are charged only for the work it does, and every job shows what it cost when it finishes.

Each effort has a research budget. A job that reaches it stops, keeps what it found and writes it up, and you can review the partial report before continuing.

Use the one-call REST API, workspace control endpoints or an HTTP MCP client. Python and TypeScript SDKs are available from the repository but are not published packages yet. The docs explain installation, authentication and the run lifecycle.

Test keys are read-only: GET, HEAD and OPTIONS. Starting research requires a live key with the required write scope and an eligible workspace plan. A test key does not become writable when a write scope is selected.

Each effort caps what a job can use. The finished run shows its actual charge; API clients use charged_micro_usd and divide by 1,000,000 to display USD. Check current plans and effort prices before estimating a workload.