Nerve Center

A cinematic operations console for Claude Code. One local screen that reads your Claude Code data and turns it into signal — usage, skills, your second brain, and a live terminal — without sending anything off your machine.

0. Operator Profile (top) — your name and account vitals (plan, member-since), plus a hookup to your second-brain folder. Drag in (or pick) a folder of notes and Nerve Center ingests them locally, then builds an interactive Knowledge Graph — a force-directed map of the concepts, files, and entities in your notes and how they connect. Hubs glow, communities are color-grouped, and you can pan / zoom / click any node. The graph is built by running /graphify headlessly (claude -p, no API key) over your notes; nothing leaves your machine. Stored in ~/.nerve-center/.
1. Terminal — a stylized, in-page bridge to a live claude session. Talk to Claude Code right inside the dashboard; it streams a headless run over your subscription (no API key), with selectable permission modes and multi-turn continuity (--resume).
2. Plan Usage — your Claude subscription usage, not API cost. Auto-detects your plan (Pro / Max 5× / Max 20× / Team / Enterprise) from Claude Code's local config, then shows two live bars — the 5-hour session window and the weekly window — with % used, % left, and a countdown to reset (using Claude's real planLimitsEndDate). Below that, token throughput over 24H / 7D / 30D / ALL (tokens, input/output, cache hit rate, sessions). No API key, nothing billed — it reads usage you already paid for in your plan.
3. Most Used Skills — your skills ranked by how often you've actually invoked them. Click any skill to open a composer ("new window") and either Run Inline (streams a headless claude -p run live) or Launch in Terminal (opens a real interactive claude session) — or just copy the command.
4. Social Signal — track your own creator follower counts for YouTube, Instagram, and TikTok. Free, no API keys — counts are fetched from your machine and cached. Enter a handle or full profile URL per platform.
5. Industry Signal — a live board of AI/agent/dev numbers pulled from Hacker News and GitHub (story points, comment counts, repo stars), refreshed on demand.

It is a local tool: everything is read from your machine. No data leaves your computer except the public Hacker News / GitHub feeds for the signal board and the public, keyless lookups used for social follower counts.

Run it

npm install      # first time only
npm run dev      # then open the printed URL (default http://localhost:3000)

Tip: set a fixed port with PORT=3737 npm run dev.

For a production build:

npm run build && npm run start

How the data works

Section	Source	Notes
Profile	~/.claude.json (oauthAccount)	Plan, account vitals, and a default name read at runtime.
Knowledge Graph	your second-brain folder → ~/.nerve-center/	Notes are ingested locally; /graphify (headless claude -p) builds the node/edge graph; rendered with a hand-rolled force simulation. Nothing leaves your machine.
Usage	~/.claude/projects/**/*.jsonl	Parses assistant records' message.usage. Incrementally cached by file mtime, so first load parses everything, later loads are instant.
Skills	~/.claude/skills, ~/.claude/plugins/**/skills, ~/.claude/commands + transcript invocations	Catalog from SKILL.md frontmatter; usage counts from Skill tool calls and <command-name> tags.
Social	public keyless lookups (per platform)	Follower counts fetched locally from your IP and cached. No API keys.
Industry	hn.algolia.com + api.github.com	Keyless public APIs, cached 5 min, graceful offline fallback.

Costs are estimates using standard Anthropic API rates (see lib/pricing.ts) — useful as a relative signal, not a billing statement.

Knowledge-graph build: the graph is built by spawning claude -p "/graphify …" --dangerously-skip-permissions on your machine, over your selected notes folder, via your logged-in claude CLI. The skipped permission prompts are expected (it runs headless) — nothing is sent anywhere, and it only ever reads the folder you point it at.

Stack

Next.js 16 (App Router) · React 19 · Tailwind v4 · Geist fonts · hand-rolled SVG chart and a dependency-free canvas force-directed graph (lib/force-sim.ts). Route handlers run on the Node.js runtime to read the filesystem and spawn the claude CLI.

Plan usage — how it's computed (and calibrating it)

Plan + reset timing are exact (read straight from ~/.claude.json: oauthAccount.organizationRateLimitTier and planLimitsEndDate). The % used is an estimate: Anthropic meters plan limits in a weighted compute unit it doesn't publish as a token count, and the live server-side percentage is only fetched by Claude Code's interactive /usage command — it isn't cached on disk.

So Nerve Center estimates consumption by weighting your local token usage the way limits actually accrue (output ≫ input ≫ cache reads) and comparing it to a per-plan budget that scales off a Pro baseline. To make the bars match what /usage shows you exactly, calibrate the budgets for your machine:

NERVE_SESSION_BUDGET=240 NERVE_WEEKLY_BUDGET=1000 PORT=3737 npm run dev

When either is set, the dashboard shows a CALIBRATED badge instead of ESTIMATED.

Config

• CLAUDE_CONFIG_DIR — override the Claude home dir (defaults to ~/.claude).
• NERVE_DATA_DIR — override where Nerve Center stores its data (defaults to ~/.nerve-center).
• NERVE_SESSION_BUDGET / NERVE_WEEKLY_BUDGET — calibrate plan-usage budgets (weighted units).
• The in-page Terminal and "Launch in Terminal" are macOS-oriented (the latter uses open -a Terminal).

---

© 2026 CortexTools. All rights reserved. This source is published for transparency and distribution; it is not licensed for reuse, redistribution, or modification.