What is Jev Search
Jev Search is open-sourced by the Search1API team AI search engine, based on TypeSafe's Jev model. Users ask questions in plain language, and Jev can automatically understand intent, select sources, limit the time range, and generate query terms. Search1API simultaneously calls 12 engines including Google, Reddit, Hacker News, arXiv, and more. Jev Search assigns each result a relevance score from 0-100 and displays them in a streaming fashion, with low-scoring results grouped separately.
Jev Search's main features
-
Natural language search: Users can ask questions directly in everyday spoken language without having to distill keywords themselves.
-
Intent understanding: The Jev model automatically analyzes user needs and determines the search topic, sources, and time range.
-
Multi-engine parallelism: The system simultaneously calls 12 search engines including Google, Reddit, Hacker News, arXiv, and more. The failure of a single engine does not affect the overall results.
-
Relevance scoring: Jev Search assigns each search result a relevance score from 0 to 100, and low-scoring results are automatically placed in a separate group.
-
Streaming display: Search results are displayed as they arrive; engines that return first are shown first, with no need to wait for all to finish.
-
Manual override: Users can directly modify source tags, time range, and query terms on the page.
-
Only finds answers, does not give answers: The system returns only links and summaries and does not generate AI summaries, avoiding fabrication and omissions.
Follow on WeChat and reply with "open source" to joinAI open-source project communication group
How to use Jev Search
- Visit the site: Visit the Jev Search official website https://jev.s1.dev/，或”添加到主屏幕”安装为 PWA app.
- Enter a question: Describe your needs in plain language in the search box, with no need to distill keywords.
- View Jev's judgment: Jev Search automatically understands intent, and the selected sources, time, and query terms are displayed on the page as tags.
- Manual adjustment (optional): Click source or time tags to add, remove, or modify them, and query terms can also be edited directly.
- Browse results: Results are displayed in a streaming fashion in the order the engines return them, and each link is annotated with a relevance score from 0-100.
- View low-scoring results: Links with lower relevance are placed in a separate group and can be expanded for viewing.
- Self-hosted deployment:Clone the repository from GitHub, configure the Search1API and Jev API Keys, and deploy to Cloudflare Workers.
Jev Search's core advantages
-
Does not generate answers: Returns only links and summaries, fundamentally avoiding the problems of AI fabrication and omission of information.
-
Search intent automation: Jev translates plain language into precise query terms, sources, and time ranges, saving users the trouble of manual filtering.
-
Transparent and controllable ranking: Each result has a relevance score from 0-100 and editable filter tags, so users can clearly see the basis for ranking and intervene at any time.
-
Broad multi-engine coverage: A single query calls 12 engines in parallel, covering web pages, communities, and vertical content, and the failure of a single engine does not affect the whole.
-
Fast response, smooth experience: Streaming output plus edge deployment means results are available as they arrive, with low global access latency.
-
Open source and self-hostable: Users can deploy their own search engine at low cost, and data is not retained by the platform.
Jev Search project address
- Project website：https://jev.s1.dev/
- GitHub repository：https://github.com/superagents-lab/jev-search
Comparison of Jev Search with similar competitors
| Comparison dimension | Jev Search | Perplexity |
|---|---|---|
| Product form | Open-source search engine frontend (MIT license) | Commercial closed-source AI search product |
| Result form | Only links + snippets,does not generate answers | Generates summary-style answers + cited links |
| Ranking method | The Jev model scores each result with a relevance score of 0-100, and the scores are visible | Internal ranking; users cannot see the basis for scoring |
| Controllability | Sources, time, and query terms can all be manually modified and overridden | The domain can be limited via Focus, but the query process cannot be intervened in |
| Search coverage | 12 engines in parallel (including HN, Reddit, arXiv, WeChat Official Accounts, etc.) | Self-built index + partner sources, with weaker coverage of community content |
| Deployment method | Can be self-hosted on Cloudflare Workers, with data under your control | Can only use the official service |
Application scenarios of Jev Search
-
Developer technical research: Search “Rust async runtimes on Hacker News this month” to quickly aggregate real community opinions rather than documentation marketing.
-
Academic paper tracking: Use “New papers on speculative decoding” to let Jev automatically lock onto arXiv and discover new papers as soon as possible.
-
Product reputation research: Search “What do Reddit users think of the Framework laptop?” to capture real user feedback from multiple communities at once.
-
Competitor and market monitoring: Limit the time range to search industry keywords, view the latest results from each engine in a stream, and keep up with market trends.
-
Multi-perspective verification of trending events: Search news, social platforms, and vertical sites in parallel for the same event, compare reports from different sources, and avoid single-source bias.