Repository navigation
Releases
NeuralDash v0.7.1
README & registry presentation update: centered hero icon & title, Open VSX / GitHub Release / MIT badges, unofficial callout, emoji feature summary, table of contents, installation instructions, and live sidebar/48h-trend preview screenshot (docs/preview.png).
NeuralDash v0.7.0
Live throughput, TTFT, cache hit %, and 3-mode sorting on the Energy Board:
- Performance metrics (
?view=performance) — parses per-model first-token latency (p50/p95), throughput (p50/p95 tok/s), 7d cache hit %, and Flex-tier dispatch wait (p50/p95) fromhttps://portal.neuralwatt.com/energy-pricing?view=performance. Missing values render as—; Flex rows never fabricate throughput. - 3-mode sort (
NeuralDash: Set Sort Order) — replaces the old sort commands with a QuickPick + board toolbar button for Bang for Buck (throughput p50 ÷ energy, default), Fastest (throughput p50desc,TTFT p50asc), and Lowest Price (energyasc, cost tiebreak), persisted inglobalState. - Mode-aware TreeView & tooltips — TreeView row descriptions lead with the active mode's primary metric; hover tooltips show the TTFT fast/mid/slow dot (
<300msgreen,<1000msyellow, else orange), throughputp50 · p95, cache hit %, Bang for Buck ratio, and the active-mode#1badge (Top Efficiency/Speed Leader/Lowest Cost). - 48h Trend Telemetry Bento & indicator tags — sidebar 48h Trend view renders a compact KPI grid with average-relative color coding (energy & throughput) and right-side value indicator tags on the SVG charts.
- Flex tier sorting, compact hover & 48h Trend — Flex rows follow the active sort mode, match the compact hover table layout, and open in the 48h Trend view with energy and chart bounds scaled by the published
flex_discount_multiplier. - Tracking alias resolution — parses
data-family-pointersalias links from?view=performanceso tracking-alias Flex rows resolve to their linked backend model's energy, 48h chart, and performance metrics, showing<alias> → <backend-slug>in the 48h Trend header. - Status bar — shows
neuraldash.myModelwhen configured, or pins the#1model under the active sort mode as⚡ <model> · <p50> tok/s · <energy> (<mode>).
NeuralDash v0.6.0
Thinking effort on hover cards, and a board fix for sub-milliwatt models:
- Thinking line — model and Flex hover cards state whether thinking is available, whether it is always on, on by default or off by default, the default effort, and the supported effort levels exactly as published. Read from
https://api.neuralwatt.com/v1/models(metadata.capabilities.reasoning,metadata.reasoning), the document NeuralDash already fetches for Flex pricing, so there is no new request or setting. Flex rows use their own catalog entry, then their base model's. Models the catalog does not list show no line.NeuralDash: Debug Tooltipshows the same line. - Same-weights note — a lane with thinking off (e.g. a fast lane) also shows the thinking of catalog entries that share its
metadata.huggingface_id, when they agree:Thinking · not available · same weights: on by default (high). - μWh energy —
https://portal.neuralwatt.com/energy-pricing?view=energypublishes very low-energy rows (an embedding model) in μWh, which NeuralDash did not parse, so those rows showed no energy. They now parse and rank, and values below 1 mWh display as μWh everywhere, with one unit per range. - No energy data section — rows the board publishes no energy for move out of the ranked list into a collapsed section at the bottom.
- The Flex hover card is now built by a pure core function, tested the same way as the model card.
NeuralDash v0.5.1
Fixes for the redesigned Neuralwatt portal:
- Board restored —
/energy-pricingnow opens on a performance view by default, which left the board with no energy values. NeuralDash now fetcheshttps://portal.neuralwatt.com/energy-pricing?view=energyexplicitly. - /models card prices — the portal's price footer changed to
$X in / $Y out per 1M tokenswith a separate cached-input row. The card parser (https://portal.neuralwatt.com/models) reads both the new and old layouts, so the Flex price fallback works again.
NeuralDash v0.5.0
Benchmarks, a sidebar chart, and redesigned hover cards:
- Live benchmark scores — rows are scored with the Epoch Capabilities Index, read at most daily from Epoch AI's benchmark download (
https://epoch.ai/data/benchmark_data.zip,eci_scores.csv, CC BY 4.0). Board models are matched by normalized slug, display name, then family; ambiguous or unknown models stayno benchmark, and fast/speed variants inherit their family score labelled(family). Value ranking (score ÷ mWh) now works out of the box. Bundled and override scores still win per slug. - 48h trend in the sidebar — the chart moved from an editor tab to a
48h Trendpane under the board. Clicking a model row (or the chart action / tooltip link) shows it there, and it re-renders on every board refresh. The chart is now an inline SVG with 7-day typical and 48h-average reference lines, hi/lo markers, per-point hover values and a low/median/high strip. - Redesigned model hover card — headline energy with the board's trend badge, a compact energy table, a text sparkline linking to the chart, the top prompt-size bands ranked by traffic share, and an
Updated … · sourcefooter. - Flex card: Normal → Flex comparison — per-token price table (input, output, cache read) and a charged-energy table (energy per request and cost per 1k requests), both computed from the
flex_discount_multiplierinhttps://api.neuralwatt.com/v1/models, which per Neuralwatt's docs applies to token billing, pay-as-you-go energy and subscription allowance drawdown. Provenance collapsed to a singleSources:line; full detail stays inNeuralDash: Debug Flex. - Board markup fixes — the energy board now groups models under family headers: rows are named from the header (
Kimi K3 Fast,DeepSeek V4 Flash · Speed lane) and keyed by their real model id; the context band is read from any sub-line segment; 48h trend direction reads the newnw-spark-cheaper/nw-spark-pricierclasses. - Cost per 1k requests shows 2 decimals; flex token prices keep up to 4 so sub-cent prices don't round to zero.
- CI: gitleaks secret scan on push and pull requests.
NeuralDash v0.4.11
First public release of NeuralDash — a live NeuralWatt energy-per-request board for VS Code.
Install
Download neuraldash-0.4.11.vsix below, then run Extensions: Install from VSIX… from the Command Palette and select the file. Works in VS Code, VSCodium, Cursor, and Windsurf.
Also being published on Open VSX: https://open-vsx.org/extension/legesus/neuraldash (pending review at time of release).
Highlights
- Live energy-pricing board: per-model mWh pricing with severity colors, 20-min polling with ETag caching and stale-first render
- Value ranking and sort by energy / value / name; status bar item for your model
- Flex tiers section with live models.dev registry refresh
- 48h trend chart panel per model (right-click a model row or use the tooltip link)
Verify
SHA-256: 3671cf3be900e4caf7d6905e59824eef01c4657f9be311f2c0d9180950de6507
Full history: CHANGELOG.md