Eonrise

Finds the research fields that already solved your problem under another name.

Features · How it works · Install · Examples · FAQ · Docs

eonrise --demo --out output

A problem that feels new in my field is often an old one somewhere else. Someone in law, medicine or materials science has usually studied it for years, under words I would never think to search for. Eonrise takes a question in plain language, searches up to six scholarly sources, and returns a short brief: what the literature says about the question, what it admits it cannot do, and which distant fields share the same shape of problem.

Features

Every term in the brief comes from a paper's title, abstract or keywords. When the papers do not support part of the map, that part stays empty and the brief says so.

How it works

Each run grows a map of concepts around your question: broader fields above it, related fields beside it, narrower ones below, each branching again. Concepts backed by the same papers are linked too. This is the written instructions question run with --max-depth 30 --max-nodes 2000: 1,950 concepts, 20 levels deep. Click the picture for the live map, where you can hover, search and switch links and colours. tools/graph_view.py builds both from any run's research_map.json.

1. The question becomes one root topic, and every later search stays tied to it.
2. Each round searches the scholarly sources and drops papers about something else.
3. Terms, relations and stated limitations come from titles and abstracts. Each term joins the map as a parent, sibling or child of the root.
4. Empty parts of the map send targeted searches back to step 2, up to --max-rounds.
5. Papers from far fields that share the question's terms are kept apart and ranked. They are the point of the tool.
6. The run writes the brief, the map and the graph, and says how much of the reachable literature it saw, so a thin map reads as unfinished.

Install

Eonrise needs Python 3.11 or newer and nothing else.

pip install eonrise
eonrise --demo --out output

The demo runs offline on a sample of 99 papers about finding analogous solutions in distant fields. It writes the brief to output/report.md, with four other output files, in about a second, and keeps no cache or stores.

Then ask your own question:

eonrise "how can written instructions prevent the same mistakes from repeating when every worker is new" --out output

Four sources run by default, with no key. Semantic Scholar and CORE join when you give their free key; docs/USAGE.md lists the environment variables.

With an AI agent

Small models tend to treat papers from distant fields as noise, which throws away the point of Eonrise. The Eonrise skill teaches an agent to run it, read the brief, and turn each distant paper into a concrete idea. In Claude Code:

/plugin marketplace add Superstos/Eonrise
/plugin install eonrise@eonrise

In Codex, the same marketplace works:

codex plugin marketplace add https://github.com/Superstos/Eonrise
codex plugin add eonrise@eonrise

Other agents can read plugin/skills/eonrise/SKILL.md directly. tools/agent_check.py measures the difference; on Haiku the skill raised the distant fields used per answer from 1.6 to 2.8 (5 runs each).

Tech stack and why

• Python standard library only. A research tool should install anywhere in one step, and zero dependencies means nothing to break, audit or pin.
• Open scholarly APIs. Crossref, OpenAlex and Europe PMC cover most fields for free and return abstracts and topic hierarchies, which is where the cross-field links come from.
• JSON and JSONL files on disk. The paper store and the synonym graph are plain files you can read, copy or delete, with no database to run.

Architecture

Examples

These are real runs, copied unedited into examples/.

What it is good for:

• Before designing something, to see which field already measured the failure you are about to hit.
• When searches in your own field keep returning the same ten papers.
• To find the words another field uses, so the next search reaches its literature.
• As the research step for an AI coding agent: research_map.json gives it sources it can cite.

FAQ

Docs

• docs/USAGE.md: bundle format, output files, environment variables, what is saved.
• examples/: real briefs from different fields.
• docs/SPEC.md: design history, benchmarks and rejected experiments, oldest first.
• CHANGELOG.md: user-facing changes and renamed names.
• CONTRIBUTING.md: checks to run and what a change should look like.

License

MIT, see LICENSE. Papers and abstracts returned by the sources keep their own licenses, so check a source's terms before sharing a saved paper store.