View .jsonl files instantly, entirely in your browser. Nothing uploads.
jsonlviewer.com — coming soon. Run it locally today (see Quickstart).
JSONL (JSON Lines / NDJSON) is the default format for LLM fine-tune data, eval logs, and agent transcripts. Every "JSON viewer" online chokes on it for the same three reasons:
- Paste fails past line 1. JSONL isn't valid JSON as a whole document — pretty-printer tools error out on the second line.
- Files are huge. Eval logs and fine-tune exports run into hundreds of MB to GB. Browser-paste tools cap out around 5MB.
- Files are private. Logs, transcripts, and training data are not things anyone should upload to a random ad-supported site to view.
JSONL Viewer is a static, framework-free web app that streams the file in a Web Worker, indexes line byte-offsets instead of parsing everything up front, and renders only what's visible. Nothing about the file ever leaves the tab — there's no server, no upload endpoint, and the shipped Content-Security-Policy enforces it (open devtools and check).
The other half: most JSONL in the wild is an LLM conversation — an OpenAI fine-tune line, an Anthropic message, a Claude Code session entry. JSONL Viewer detects that shape per line and renders it as a readable chat transcript instead of a wall of escaped \n characters, with tool calls and thinking blocks collapsed by default.
- Streams gigabyte files. A Web Worker scans the file once for line boundaries and stores byte offsets — it never holds the whole parsed file in memory. Row content is fetched lazily as you scroll.
- Auto-detected transcript view. Recognizes OpenAI
{messages:[...]}, Anthropic message/content-block shapes, and Claude Code session lines — one line at a time, since real files mix formats. See docs/DETECTION.md for the exact rules, including how to add a new format. - Search: text or fields. Plain text is a live substring filter. Field queries like
role:assistant,tokens>500, orusage.total_tokens<9run in the worker with cancellation, so a fast retype never queues stale work. - Nothing leaves the browser. No server, no analytics on file content, no external network calls.
public/_headersships a CSP with no external origins — verifiable in your network tab. - 38KB gzipped, dependency-free framework, virtualized rendering — the whole app loads faster than most single hero images.
| JSON mode | Transcript mode |
|---|---|
 |
 |
| Light theme |
|---|
 |
git clone https://github.com/viswesh/jsonl-viewer.git
cd jsonl-viewer
npm ci
npm run devOpen the printed local URL, drop a .jsonl file (or click try a demo file), and it renders immediately — no build step required for local use, no account, no upload.
npm run dev # local dev server
npm test # unit tests (Vitest)
npm run e2e # browser smoke tests (Playwright)
npm run build && npm run check-size # production build + bundle budget gate (<50KB gz)See CONTRIBUTING.md for the full workflow, code style, and PR checklist.
No framework — Vite + vanilla TypeScript. A Web Worker owns file I/O, line indexing, and search; the main thread only renders. See docs/ARCHITECTURE.md for the full design: data flow, the virtualized list, the sanitization boundary, and why each dependency (marked, dompurify, @speed-highlight/core) is the only three the project takes.
File content is treated as fully untrusted input. Every markdown-rendered string passes through DOMPurify before touching the DOM; JSON values render via textContent, never innerHTML. See SECURITY.md for the threat model and how to report a vulnerability.
Bug reports, feature requests, and PRs are welcome — see CONTRIBUTING.md. This project follows the Contributor Covenant.