AURA — the dependency-free token saver (CLI · MCP · library)

Cut your LLM bill by answering recurring prompts for free — from a local cache, saved skills, or deterministic compute — before you ever call a model. Use it in your terminal, drop it into code, or wire it into Claude / Cursor / Claude Code via its built-in MCP server. Zero dependencies, MIT licensed.

AURA answers your prompts the cheapest way first so you call (and pay for) an AI model far less:

1. Cache (exact) — you asked this before → instant, free.

2. Cache (fuzzy) — you asked something similar → free. (Filler words like "what's the…" are ignored, so close rephrasings still hit.)

3. Compute — solved locally, free: math, unit conversions, dates, base64, word-count, upper/lowercase, percent of, % off (with savings), tip, percent change, days between dates.

4. LLM fallback (optional) — only if you add --llm and you have a model key set. AURA auto-picks the cheapest capable model (light / balanced / heavy) for the question, then caches the answer so next time it's free.

Inspired by AINL (AI Native Lang). AINL's core idea is to keep the model off the hot path: figure something out once, then run it deterministically forever with no per-run inference. AURA applies that same principle — cache + local "templates" mean recurring questions cost nothing.

It uses only built-in Node — no installs, no dependencies. Cache lives in ~/.shaddai-aura (your home folder), so it works from any terminal.

Use it (no install)

Open a terminal in this folder and run:

node cli.js ask "what is 15 * 240"
node cli.js ask "convert 10 km to miles"
node cli.js stats

Related MCP server: AgentLayer MCP Server

Use it from anywhere (type aura instead of node cli.js)

Run this once, inside this folder:

npm install -g .

Now from any terminal:

aura ask "what is 12% of 80"
aura learn "our refund policy" "30 days, no questions asked"
aura ask "our refund policy"          # → free, from cache
aura stats

(To undo the global install: npm uninstall -g @shaddai/aura.)

Commands

Saved skills (define once → free forever)

A skill is a tiny "compiled program": a pattern → a deterministic action, stored in ~/.shaddai-aura/skills.json. Once saved, any matching prompt is answered for free with no AI call — AURA's take on AINL's "author once, run forever."

# substring/keyword match → fixed answer
aura skill add "support" --match "support email" --do "cloudzncrownz@gmail.com"
aura ask "hey whats the support email"     # → cloudzncrownz@gmail.com   (free · via skill)

# regex match with $1, $2 capture-group substitution (--regex, or wrap the pattern in /.../)
aura skill add "greet" --match "/^hi (\w+)/i" --do "Hello, $1!" --regex
aura ask "hi Brittany"                      # → Hello, Brittany!

aura skill list                             # show all saved skills
aura skill remove "greet"                   # delete one

Matching: a plain pattern matches if every word in it appears in the prompt (case-insensitive). Wrap it in /.../ (or pass --regex) to use a regular expression; capture groups fill $1, $2, … in the answer.

Where skills sit in the flow: route() checks exact cache → fuzzy cache → skills → local compute. A real cached answer wins (it's the most authoritative), but your explicit skill beats generic compute.

Adapters (live data, still free, no key): a skill action can be { type:'adapter', adapter:'price', args:{ coin:'btc' } } to fetch deterministic data instead of calling an LLM. The built-in price adapter uses CoinLore (no API key). Adapters do network I/O, so they run through the async ask() (not the sync route()) and degrade gracefully — if you're offline they just return a normal miss, never an error. Define one in JS:

const aura = require('./aura-core');
aura.addSkill({ name: 'btc', match: 'btc price', action: { type: 'adapter', adapter: 'price', args: { coin: 'btc' } } });
await aura.ask('btc price');   // { method:'skill', answer:'Bitcoin (BTC): $65785.37', ... }

Connecting your AI model (for --llm)

Set one of these before running (whichever service you have a key for):

# Windows PowerShell
$env:OPENROUTER_API_KEY = "sk-..."     # or OPENAI_API_KEY, or ANTHROPIC_API_KEY

# macOS/Linux
export OPENROUTER_API_KEY="sk-..."

Then aura ask "summarize this..." --llm works. Without a key, --llm simply tells you no model is connected — it never makes anything up.

Use it in Claude / Cursor / Claude Code (MCP)

AURA ships an MCP server so any Model Context Protocol client can save tokens automatically. It exposes three tools:

The server is zero-dependency — it speaks MCP's JSON-RPC over stdio directly.

Claude Desktop

Add to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "aura": { "command": "npx", "args": ["-y", "-p", "@shaddai/aura", "aura-mcp"] }
  }
}

Claude Code

claude mcp add aura -- npx -y -p @shaddai/aura aura-mcp

Cursor

Add to .cursor/mcp.json:

{ "mcpServers": { "aura": { "command": "npx", "args": ["-y", "-p", "@shaddai/aura", "aura-mcp"] } } }

Running from a local clone (before the npm package is published)

Point the client's command at your checkout instead:

{ "mcpServers": { "aura": { "command": "node", "args": ["/absolute/path/to/aura/mcp.js"] } } }

How it saves tokens: tell your assistant (system prompt / project rules) to call aura_ask before answering, use the answer if hit is true, and call aura_remember after generating a stable answer. Recurring questions, facts, and anything computable then cost nothing.

Use it as a library (for the dashboard / other code)

const aura = require('@shaddai/aura');     // or require('./aura-core')
const r = aura.route('what is 2+2');        // { hit:true, method:'compute', answer:'4', ... }
const full = await aura.ask('...', { llm: true });
aura.recordAnswer('q', 'a');                // cache an answer
aura.stats();                               // savings summary

// saved-skills registry
aura.addSkill({ name:'support', match:'support email', action:{ type:'answer', text:'...' } });
aura.listSkills();
aura.matchSkill('whats the support email'); // { name, action, text, ... } or null
aura.removeSkill('support');

This is the same engine as the SHADDAI dashboard's backend/lib/aura.js.

Share one cache between the terminal tool and the dashboard

Both the CLI and the dashboard read AURA_HOME if it's set. Point them at the same folder and their savings compound — an answer learned in the terminal is free in the app, and vice-versa:

# Windows PowerShell (set once for your user)
setx AURA_HOME "$env:USERPROFILE\.shaddai-aura"

If AURA_HOME is unset, the CLI uses ~/.shaddai-aura and the dashboard uses its own backend/data folder (separate caches).