TVAKSECURITYDOCS
AvailableGetting Started

Product Architecture

How evidence moves from an endpoint to an investigation.

Acquisition and analysis are separate

TVAK is split into two parts. Collector is thin, fast, portable evidence acquisition only, and runs directly on (or against) the target machine. Analyst is a separate product, in development: the planned investigation console for processing what Collector produces, offline, away from the target.

Collector's only job is acquisition. It never makes a forensic judgment about what it collects. Filtering, scoring, and correlation happen later, offline, in TVAK Analyst.

Windows endpoint          .zip / .tvak package            analyst workstation
Collector (GUI or CLI) -> manifest, SHA-256, log   ->     TVAK Analyst (in development)
no permanent agent

Two front-ends, one engine

The GUI (TvakCollectorGui.exe) and the CLI (TvakCollector.exe) drive the exact same collection engine, so a GUI run and a headless run of the same selection produce identical results. See GUI and CLI.

Evidence modules

Collector gathers evidence through modules, selected with --modules on the CLI or the checklist in the GUI. Processes is always collected. Each module that produces output gets its own subfolder under artifacts/ in the package. Module by module detail is in Evidence Handling.

The evidence package

Every collection is written as a single self-identifying package. Both extensions, .zip and .tvak, are the same ZIP-based container.

<name>.zip  (or .tvak)
├── tvak_format.json     format identity marker, written first
├── manifest.json        per-file SHA-256, collector run status, host/collector provenance
├── metadata.json        case-level context, empty by default
├── collection.log       full run log
└── artifacts/           one subfolder per module that produced output

Every file is SHA-256 hashed and recorded in manifest.json, together with a per-collector run status. Field by field detail is in File Formats.

Where analysis runs

Core evidence processing is designed to occur locally by default, on the analyst workstation. External services, including AI APIs, remain optional and must be explicitly configured. Analyst is in development; this describes its design, not a released feature.