nodecode-machine-library
Machine Library: research search over papers, books, patents and standards, with citable source links. A Nodecode rewrite of SpaceFrontiers/hermes-machine-library by SpaceFrontiers.
Get it
Tell the agent "add the machine-library cell", or run nodecode add nc://machine-library.
Set it up
The cell talks to Machine Library's API with a key from machinelibrary.ai/keys. Searches spend the account's prepaid credits. Either export it:
export MACHINELIBRARY_API_KEY=...
or set it in /settings (or the page's Settings tab) under
machine-library: api_key for the key itself, or api_key_env for the
name of the variable holding it. A key pasted into the panel is kept in a
0600 file under ~/.nodecode/secrets/; the folder holds no secret.
Nothing else is required. /machine-library with no argument shows which
endpoint and key the cell is using, and (help :machine-library) is the
model's guide to the tools.
Use it
Ask for what you want; the model picks the tool:
| Tool | Use |
|---|---|
machine_library_search |
find documents — title, page url, byline, the matching passage, and the identifier to cite |
machine_library_extract |
read a document the library holds: its /r/... page, or a DOI, arXiv, ISBN or standard identifier |
From the keyboard, /machine-library <query> runs one search and shows its
sources.
A search answers with a https://machinelibrary.ai/r/... page for each
document and the identifier behind it — doi://10.48550/arXiv.1706.03762,
iso://1219-1:2012, isbn://... — so an answer can cite where a claim
came from. Extract reads what the library stores for a document. A url the
library does not hold fails on its own line: pair the cell with a general
page fetch (the shipped websearch cell fetches the open web) for anything
else.
Every call spends the account's credits, and max_chars bounds how much of
one document a tool result carries.
The two tools are the plugin's two verbs
The original registered a web search provider with Hermes, so that
web_search and web_extract asked Machine Library. A cell registers
tools of its own instead, so the same two calls are the two tools above,
against the same API and with the same request shapes: POST /v2/search/
for search, and GET /v2/documents/by-uri/... or
GET /v2/documents/{id} for extract.
Not ported
- The provider plug-in point itself. Hermes let a plugin take over its
web_searchandweb_extractverbs; Nodecode's web search is its own cell, so this cell registers two tools rather than replacing a verb. - The
web.backendselection in Hermes's config. Here the section is the switch: a presentmachine-librarysection is an enabled cell. - The search index choice. The provider sent no index selection either, so nothing is lost; the API's own default profile answers.
Test
From the cells repository root:
.github/test-cell.sh tools/nodecode-machine-library
Every request is a stubbed nlk:http, so nothing reaches Machine Library
and no key is needed.
MIT licensed, as the original is.
