nc://jina

Jina Reader: any web page as clean Markdown, and Jina Search's sources, as two tools the model calls. A Nodecode rewrite of Adolanium/hermes-jina by Adolanium.

Get it

Tell the agent "add the jina cell", or run nodecode add nc://jina.

Set it up

Reading works with no key at all, at Jina's anonymous limit of roughly 20 requests a minute. Searching needs one: make it at jina.ai/api-dashboard and either export it:

export JINA_API_KEY=jina_...

or set it in /settings (or the page's Settings tab) under jina: 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. /jina with no argument shows what the cell is using and whether a key is set, and (help :jina) is the model's guide to the tools.

Use it

From the keyboard, /jina <url> reads that page and /jina <words> runs one search.

One broken url is its own error under the pages that did read, as the plugin answered a batch; a url that is not http:// or https:// is refused before a request is made. max_pages, max_chars, max_results and timeout_seconds in the section bound what one call costs.

Jina fetches the page on its own servers, so the urls the model reads and the words it searches go to r.jina.ai and s.jina.ai. The key rides as a bearer token to Jina and nowhere else, and an error names the status and Jina's own message, never the key. That is the whole of what leaves this machine.

The two tools are the plugin's two operations

The original registered a web backend with its harness, so the harness' own web_extract and web_search reached Jina through it and it registered no tool of its own. Nodecode has no such switchboard, so the backend's two operations — extract and search — are two tools here, posting to the same two endpoints with the same headers.

Not ported

• The backend switch. In the plugin the operator chose Jina per operation, in the harness' web_extract / web_search settings. A Nodecode cell is present or absent, so once the folder is on, the model sees jina_read and jina_search beside whatever other web cell is on.
• Parallel reading. The plugin read several urls at once, 2 in flight without a key and 5 with one, to stay under Jina's rate limit. This cell reads them one after another, at most max_pages a call.
• The harness' own web-tool plumbing — per-profile secret scopes, the backend's is_available gate and its setup-schema rendering. The section's guide, the /jina status line and the cell's start notice say the same things here.

Test

From the cells repository root:

.github/test-cell.sh tools/nodecode-jina

Every request is a stubbed nlk:http, so nothing reaches Jina and no key is needed.

MIT licensed, as the original is.