Transcript Lens
September 20, 2026 · View on GitHub
Next.js app (Türkçe UI) that explores YouTube transcripts by meaning: TypeSafe Jev classifies blocks for kind, value, and signals without rewriting the text.
| At a glance | Details |
|---|---|
| Source | Source |
| Tags | Open source · Free source build · BYOK |
| Product homepage | Project homepage — source-built Next.js app; no separate hosted product required. |
| Pricing and access | MIT source has no app purchase fee, checked 2026-09-20. Analysis needs your Vercel AI Gateway key (JEV_PROVIDER=gateway) or TypeSafe key (JEV_PROVIDER=typesafe). Provider usage can incur charges. Keyless mode still opens/pastes/exports sample text without analysis. |
| Jev evidence | Inspected src/lib/jev.ts and src/lib/analyze.ts: Gateway experimental_evaluate with model typesafe-ai/jev, or direct POST https://api.typesafe.ai/v1/systemone; per-block Choice/Noul suite (kind, value, surprising, funny, …). Live analysis not run. |
| Disclosure | Free source access does not include Gateway/TypeSafe usage. AI-assisted, independently curated listing; no commercial relationship was declared. Inclusion is not endorsement. Implementation reviewed from public source. |
| Maintainer | sensahin. |
| Format | Next.js application (transcript-lens 0.1.0) with API routes for analyze/lens. |
| Platform and availability | Source build with Node.js 22+; npm run dev on localhost, or deploy your own Vercel fork. Experimental personal tool—upstream warns open deployments share your provider bill. |
| Jev's role | Classifies transcript blocks and powers semantic search views; application code owns captions, heatmaps, sectioning, and exports. Claims are not fact-checked. |
| Requirements | Node.js 22+, npm; optional AI_GATEWAY_API_KEY or TYPESAFE_API_KEY. |
| License | MIT. |
When to use
Use it to skim long YouTube captions for substance, sponsors, and topical passages with a Turkish-first UI. Prefer Sponsor Skip when you only need sponsor-read boundaries for playback skip, or Jevmeter for video caption gauges from sentence judgments.
How it works
Transcript text is split into blocks. analyze.ts issues one multi-question Jev call per block (kind Choice, value Choice, several Noul signals) through Gateway or direct TypeSafe. Code aggregates answers into heatmaps, section titles from native phrases, and search hits that show original wording rather than summaries.
Get started
git clone https://github.com/sensahin/transcript-lens.git
cd transcript-lens
git checkout 9183ef9ff0f9a18834bb139d01e508c22d422257
npm ci
cp .env.example .env.local
npm run dev
Open http://localhost:3000, try the built-in Turkish sample transcript. Analysis without a key shows an explanatory message. Live analysis sends transcript blocks to your chosen provider and can incur charges. This listing did not run npm test or live Jev.
Examples and demos
- In-app Turkish sample transcript (project-authored; not taken from another video).
- Offline
tests/core.test.ts— not executed on the review host.
Limits and data handling
Transcript text leaves the host for Gateway or TypeSafe when analysis runs. Keys must not use NEXT_PUBLIC_ prefixes. Public Vercel deploys without auth can spend the deployer's quota—upstream documents this risk. UI language is Turkish; analysis criteria in code are English.
Review and maintenance
Reviewed on 2026-09-20 at commit 9183ef9: 0.1.0, MIT. AI-assisted source review of README, LICENSE, src/lib/jev.ts, src/lib/analyze.ts, and package metadata. No live TypeSafe/Gateway analysis.
Related: Sponsor Skip, Jevmeter.