nc://neuralwatt

v0.1.0Pinned
ThomSharktools/nodecode-neuralwatt: Read-only NeuralWatt views as nine tools: health, prices, balance, spend, energy, usage, requests, sessions. - a rewrite of 0xGingi/neuralwatt-hermes (MIT), credited to 0xGingi5835b21· 10 Nov1 commit
test1 filecell.lisp41.0 KB

nodecode-neuralwatt

NeuralWatt: your account's health, model prices, balance, spend, energy, usage, requests and sessions, as nine tools the model can call. A Nodecode rewrite of 0xGingi/neuralwatt-hermes by 0xGingi.

Every view is one or more plain GETs to the NeuralWatt API with the account's own key, and nothing is written anywhere. The API is content-free: it stores no prompt or completion text, so what comes back is counts, costs, timings, model names and session ids.

Get it

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

Set it up

One key. Paste a NeuralWatt API key into the section's api_key, or export the variable api_key_env names (NEURALWATT_API_KEY by default) and leave api_key empty. With neither, the cell refuses at start and says so.

Member Default What it is
api_key empty the key, written in the section
api_key_env NEURALWATT_API_KEY the variable holding the key, read when api_key is empty
base_url https://api.neuralwatt.com where the API answers; an https:// URL, since the key goes wherever it points
timeout_seconds 30 the ceiling on one request

A gateway of your own in front of the API is what base_url is for: that host then receives the key.

Use it

Ask, and the model calls the tool:

Tool What it answers
neuralwatt_health the gateway's own status, and the detailed component checks
neuralwatt_models the catalog: public models, and the account's own
neuralwatt_prices input, output and cached input, USD per million tokens
neuralwatt_quota balance, allowance, usage totals, rate-limit tier
neuralwatt_dashboard balance, allowance, today's spend, recent energy (days)
neuralwatt_energy kWh, joules and carbon, with a daily breakdown
neuralwatt_usage summary, by_model, by_key, or a cost and latency compare
neuralwatt_requests a page of request rows, or top_expensive
neuralwatt_sessions list, one session's detail, or its families

The same views in the shell:

/neuralwatt              the dashboard
/neuralwatt prices       the price table
/neuralwatt requests     the newest request rows

And through eval: (neuralwatt:dashboard), (neuralwatt:prices), (neuralwatt:usage-view "compare"), (neuralwatt:energy :start-date "2026-01-01"), (neuralwatt:sessions-view "detail" :session-id "...").

A comparison and a ranking scan the per-request rows a page at a time (200 rows a page, max_pages of them) and say when the window is only the pages they read, so a partial answer never reads as a whole one.

Run .github/test-cell.sh tools/nodecode-neuralwatt from the cells repository. The tests are offline: every exchange is stubbed.

Add nc://neuralwatt

  1. 1

    Install Nodecode

    Once per machine.

    $ curl -fsSL https://nodecode.ai/install.sh | sh
  2. 2

    Add the cell

    Pinned at 5835b21.

    $ nodecode add nc://neuralwatt

    Or tell the agent add the neuralwatt cell.