evvl

Evvl — Side-by-Side AI Model Comparison

Compare AI model outputs and get a shareable URL with results.

API URL: https://app.evvl.ai

Auth Check

Before anything else, find the Evvl token:

1. Check for .evvl/config in the current project directory
2. Check for ~/.evvl/config in the user's home directory
3. Check the EVVL_TOKEN environment variable

The config file is JSON: { "token": "..." }

If a token is found, skip to Workflow below.

If no token is found, run the Device Code Auth flow.

Device Code Auth

1. Tell the user you need to connect their Evvl account.
2. Try to open the auth page in their browser:

   open "https://app.evvl.ai/auth" 2>/dev/null || xdg-open "https://app.evvl.ai/auth" 2>/dev/null
   

   If that fails, tell them: "Open this URL in your browser: https://app.evvl.ai/auth"
3. Tell the user: "Sign in with OpenRouter, then paste the code you see on screen."
4. Wait for the user to paste a code (format: EVVL-XXXXXX).
5. Exchange the code for a token:

   curl -s -X POST "https://app.evvl.ai/api/auth/device/exchange" \
     -H "Content-Type: application/json" \
     -d '{"code": "<THE_CODE>"}'
   

   Response: { "token": "...", "userId": "..." }
6. Save the token to ~/.evvl/config:

   { "token": "<token>" }
   

   Create the ~/.evvl/ directory if it doesn't exist.
7. Confirm: "Connected! Your token is saved. Running your comparison now..."

Then continue to Workflow.

Workflow

Step 1 — Parse input

Extract from the user's message:

• Model names (e.g., "GPT-4o vs Claude Sonnet")
• The prompt/task (e.g., "write FizzBuzz in Python")
• Optional baseline preference (default: none)

If the message is ambiguous, ask for clarification before proceeding.

Step 2 — Resolve models

Fetch the full model list from GET https://app.evvl.ai/api/models. Use your own intelligence to match the user's request to model IDs (e.g., "latest ChatGPT" → openai/gpt-4o).

Confirm your picks with the user — wrong model = wasted money:

Models I'll compare:
- openai/gpt-4o (GPT-4o)
- anthropic/claude-sonnet-4 (Claude Sonnet 4)

Ok to proceed?

Step 3 — Run comparison

Read the token from the config file found in the Auth Check step. Auto-generate a title from the context — no need to ask the user.

curl -s -X POST "https://app.evvl.ai/api/compare" \
  -H "Authorization: Bearer <TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": { "name": "<auto-generated title>", "content": "<the prompt>" },
    "models": [
      { "provider": "openrouter", "model": "<model-id>", "displayName": "<name>" }
    ],
    "baseline": "none",
    "title": "<auto-generated title>"
  }'

Step 4 — Present results

Show:

1. Share URL (prominent)
2. Per-model summary — latency and token counts