Technical docs
Architecture
How the product layers fit together without turning the website into an execution surface.
Runtime layers
TraceLodger is backed by a Python evidence pipeline with two review shells: the stable CLI plus Streamlit review UI, and the native Tauri desktop workspace that supervises the Python backend through a local sidecar.
- Evidence target: mounted or extracted local directory tree.
- Python engine: discovery, parsing, workflows, reporting, validation, provider exports, and investigation review.
- Generated case directory: JSON, JSONL, Markdown, validation, review, and provenance artifacts.
- Review shells: Streamlit reference UI and native desktop app.
- Desktop sidecar: authenticated stdio JSONL between Rust host and Python service.
Website boundary
The public website should stay static-first. It should use curated content objects, sanitized screenshots, synthetic demo summaries, and product copy from reviewed docs.
- Do not run the local evidence engine from the public website.
- Do not publish raw evidence directories, raw local databases, complete private messages, secrets, or real identifiers.
- Interactive demos should be simulated from sanitized static content unless a separate redacted product demo environment exists.
Trust architecture
Trust pages can explain local-first evidence review, read-only source posture, generated case outputs, evidence-root containment, offline-capable desktop direction, host-supervised sidecar operation, diagnostics boundaries, and review-only mode.