exai

CLI that turns natural language into Excalidraw flowcharts. Point it at your codebase, describe what you want, get a .excalidraw file.

exai ai "visualize the auth flow" -c ./src -o auth.excalidraw

Install

npm i -g exai

Or from source:

git clone https://github.com/KailasMahavarkar/exai.git
cd exai
npm install && npm run bundle

Setup

Get an API key from openrouter.ai/keys, then either:

export OPENROUTER_API_KEY="sk-or-v1-..."

Or use a config file:

exai init                    # creates exai.config.json
# add your apiKey, then:
exai ai "prompt" --config-path ./exai.config.json

Config

One file controls everything. Generate it with exai init.

{
  "model": "moonshotai/kimi-k2.5",
  "filterModel": "moonshotai/kimi-k2.5",
  "apiKey": "sk-or-v1-...",
  "temperature": 0,

  "format": "dsl",
  "output": "flowchart.excalidraw",
  "direction": "TB",
  "spacing": 50,

  "context": ["./src"],
  "exclude": ["dist", "coverage", "*.lock"],
  "allowTestFiles": false,
  "maxFileSize": 65536,
  "maxDepth": 6,
  "maxTreeItems": 1000,

  "compress": true,
  "compressMode": "balanced",
  "compressOptions": {
    "removeComments": true,
    "minifyWhitespace": true,
    "extractSignaturesOnly": false,
    "maxFileLines": 100,
    "preserveImports": true,
    "preserveExports": true,
    "preserveTypes": true,
    "preserveFunctionSignatures": true
  },

  "cache": true,
  "cacheTtlDays": 7,
  "cacheMaxEntries": 100,

  "verbose": false
}

All fields optional. CLI flags override config values.

Commands

# AI generation
exai ai "<prompt>" [options]

# Create from DSL/JSON/DOT
exai create [input] [options]

# Parse without generating
exai parse <input>

# Cache management
exai cache stats
exai cache clear

# Generate starter config
exai init [path]

DSL Syntax

(Start) -> [Process] -> {Decision?}
{Decision?} -> "yes" -> [[Save to DB]] -> (End)
{Decision?} -> "no" -> [Retry] -> (Start)

[rect] {diamond} (ellipse) [[database]] -> --> -> "label" ->

License

MIT