# Jevlin: SuperSearch for coding agents Source: https://jevlin.ai/ · Docs: https://docs.jevlin.ai/ Jevlin How it works Tiers Benchmark Docs GitHub Get API key SuperSearch asks every engine at once, says which ones found each result, and gives you the receipt. Jevlin is SuperSearch for coding agents: eight engines in one call, free for agents, with the trajectory behind every search. # Agents Shouldn't Pay to Search. Jevlin for Search Web search for coding agents Install For Claude Code, Cursor, Codex, opencode, Pi and Hermes. One install, every agent. SuperSearch asks every engine at once, says which ones found each result, and gives you the receipt. Jevlin is SuperSearch for coding agents. One call. Eight engines. One query 01 / SuperSearch ## Introducing SuperSearch SuperSearch is one call that asks every engine at once and returns one list, each result tagged with the engines that found it, with a receipt per engine. It finds what no single engine finds alone. Every search engine has blind spots, and they don't overlap. Exa's neural index, Brave's crawl, Serper's, SerpAPI's, Perplexity's answers, Parallel's research: each one is built for something different and sees a different web. Wire your agent to one of them and it inherits that index's gaps on every question it ever asks. Rephrasing doesn't fix that. A second opinion does. So does an eighth. 02 / How it works ## How a search runs - 01 ### Ask once. Your agent runs one search. Leave the tier to the one you set, or ask for balanced when it needs several sources; pin a domain filter or a recency if you want. - 02 ### Eight answers. Every engine in your tier runs at the same time, under a per-query cost ceiling. If one fails, the rest still land. fast asks three indexes and answers in about a second. - 03 ### One response. Merged, de-duplicated by page, ordered, each page marked with the engines that found it. Titles, links and excerpts, ready for a model to read. - 04 ### Full receipt. Per-engine status, latency and cost, kept against the request id, and the search placed in its session's trajectory on the platform. 03 / What changes ## One call, eight engines, one list ### Breadth without the loop. One call fans out to every engine in your tier, in parallel. Results come back merged, de-duplicated by page, each one labelled with the engines that found it. Agreement between independent indexes is the evidence. ### Stop paying for the retry. When the first result set is thin, a single-index agent searches again, and every extra turn costs model tokens. Fan-out front-loads the coverage so your agent has less reason to go back. ### One key. No bill. Eight engines. No accounts to juggle and nothing to top up. Jevlin holds the engine keys; setup gets yours, with no browser and nothing to paste. ### Search you can audit. Every search is a trace: which engine returned what, how fast, and what it cost. Every session is a trajectory: what the agent was working on when it searched, what came back, and what it concluded. On the platform, turn by turn. 04 / Tiers ## Tiers and pricing A tier is how wide a search fans out. What a search costs depends on what it carries: with its trajectory it is free up to a daily allowance, and without one it is list price. fast ### Three indexes, about a second. - What runs: One fast index from each group, together - Results: One page of results from each - For: A routine lookup; the default for a new key - Allowance: Free, up to 200 searches a day, then $15 per 1,000 balanced ### The full core set, attributed. - What runs: Eight indexes, in parallel - Results: 8 per engine, a 12-page window, every page marked with who found it - For: More than one source, or a thin first answer - Allowance: Free, up to 50 searches a day, then $35 per 1,000 max ### Everything, for corroboration. - What runs: The same eight indexes, in parallel - Results: 15 per engine, a 24-page window, 4,000-character excerpts - For: The same page found by independent indexes. Yours to choose; the agent never sends it - Allowance: Coming soon. Contact us (https://platform.jevlin.ai/contact-us) Every tier, once your agent is claimed Rate and concurrency come with the account, not the tier. The trajectory is what makes a search free: the assistant text your agent's hook sends with it. A search from a shell with JEVLIN_TRACE=off, or from a host whose hook carries no text, is list price. Free searches count per account, up to $2 of engine cost a day. A key whose trajectories do not verify is moved to list price for a day, twice in a week until support restores it; the keys page says so. The pricing page (https://docs.jevlin.ai/use/pricing) has every number. - Searches a minute: 120 - Searches an hour: 3,000 - At a time: 16 - Before the claim: fast only10 a minute, 120 an hour, 2 at a time ### Who sets the tier You do, ahead of time: /jevlin tier balanced in the agent, or jevlin agents tier from a shell, sets the tier every search carries. The skill tells the agent to leave it alone so your tier applies, and to ask for balanced when it needs several sources. With no tier set, Jevlin applies the default for your key. A tier is also a per-query cost ceiling: a set that would exceed it is trimmed, never refused. Naming your own engines needs balanced or max, runs exactly the arms you name on your credit, and earns no reward. 05 / Built for agents ## Built for agents Install ### One install, every agent. Setup finds the coding agents on your machine and writes a skill and hooks into each: Claude Code, Cursor, Codex, opencode, Pi and Hermes. Any other agent gets a rules block to paste. npm install -g @jevlinsearch/jevlin jevlin setup Key ### No key to paste. Setup registers itself with the search platform and receives a key, with no browser and nobody at a keyboard. Search works from that moment. A person claims the agent later to attach it to an account at platform.jevlin.ai, which is also where you get a key for direct API calls. In the agent ### Yours to switch, yours to widen. The skill makes Jevlin the default search. Two commands keep it in your hands. /jevlin off — the agent's own search is the default again /jevlin tier balanced — the tier every search carries Sources ### Plain JSON. No middleman answers. SuperSearch gives your model the sources, not a paragraph written by someone else's model. Your agent reads the web, not a summary of it. Privacy ### Queries off the command line. The query travels as JSON on stdin, never in argv. The context that rides with a search is scrubbed of credentials, key blocks, email addresses and your account name before it leaves, and JEVLIN_TRACE=off sends none of it. Free ### Free for agents, no card. Setup gets a key with no card and nothing to top up. The fast and balanced tiers come with a free daily allowance; max draws on credit you add on the platform. 06 / Benchmark ## Benchmark results, published in full We benchmark on BrowseComp with OpenRouter's own prompts and graders, unmodified, and we publish every run, including the ones we lost. The harness, the methodology and the transcripts are in the open. Interim, 17 BrowseComp questions: with fan-out the agent answered 12. Alone, it answered 2. Full 50-question run and the public repo to follow. The number above is labelled interim in the methodology and will change. 07 / Start ## Install in thirty seconds Install, run setup, restart your agent. Setup gets the key for you. The quickstart has the native installer and every agent; calling the API yourself starts at how an agent gets an API key. Or let your agent do it: https://jevlin.ai/agents.md is the same install written for an agent to follow, and the Jevlin entry on skills.sh, Claude Code's plugin directory, Codex's plugin marketplace and Cursor's runs it. npm install -g @jevlinsearch/jevlin jevlin setup # or, without Node curl -fsSL https://jevlin.ai/install.sh | sh Jevlin Benchmark Docs GitHub X LinkedIn