super-research
Toy research agent where Jev makes every keep/skip and delve decision over search results, and one expensive call writes the report.
Overview
super-research is a self-described vibe-coded toy for an idea: a research agent that spends on just one expensive model call per pass. Cheap components do the routing: a flash LLM drafts seed queries, a small local model (Needle) extracts named concepts from pages, Tavily and SearXNG search, and Jev makes every keep or skip decision. A single OpenCode Go model call then writes the report from a knowledge tree of everything the pass found. It runs as a CLI with a live dashboard or a web UI where you supply your own keys, which stay in the browser and in memory for the run only.
Jev is used at several stages, each as Nouls: one per query (run or skip), one per search result bundled per query (scrape or skip), a page-content Noul as the stop signal plus one per anchor (delve or not, recursing by depth), and a gate per kind for concept expansion that spawns new queries. A priority queue over the tree uses Jev score times depth decay, every Jev call is logged with probabilities, tokens and latency, and a success metric tracks the share of approved delves whose content Jev then scored relevant (target 0.7 or higher). Tests run offline with stubs.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Should this query be run? | — |
| Yes / no | Should this search result be scraped? | — |
| Yes / no | Is this page relevant (stop signal), and should this link be followed? | — |
How it works
- Draft seed queries cheaply
- Jev gates queries, results, pages and links
- Grow a knowledge tree with concept expansion
- One expensive model call writes the cited report
Strengths
- Most decisions are cheap Jev calls
- Every verdict logged for debugging
- Budgets and gates configurable per run
Caveats
- Toy project by the author's own description
- Needs several API keys
Getting started
Set TYPESAFE_API_KEY and OPENCODE_API_KEY (Tavily optional), then uv run research "topic".At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- CLI + web UI
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
A research agent that spends on one expensive model call per pass.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Whether to run each query, scrape each result, accept each page and follow each link.
How many expensive calls are made?
One report-writing call per pass.
Is it production-ready?
No; the author calls it a vibe-coded toy example.
Related projects
Depth-crawling research agent where Jev makes every per-page relevance and next-link decision and an LLM only plans and writes.
ProjectprototypeJavaScript- jev-social★ 169
Local social research agent: Jev picks each bounded Instagram, TikTok or LinkedIn step and socai CLI runs it in your Chrome.
CLIbetaTypeScript Research agent where GPT plans and Jev finds evidence in parallel across document regions with Choice and Noul.
Browser extensionresearchTypeScriptWeb search CLI for agents that has Jev score results and scraped chunks against the query and goal, passing only relevant ones on.
BenchmarkbetaGoMulti-agent orchestration layer for Claude Code with opt-in Jev hooks for fast judgment points, decision caching and egress controls.
PluginproductionTypeScript- winnow★ 110
Calibrated context sieve for Claude Code: Jev judges each block of large tool output and hides confident-no blocks behind recallable stubs.
Projectbeta