Skip to content

QUICKSTART.md

Latest commit

 

History

History
102 lines (72 loc) · 2.5 KB

QUICKSTART.md

File metadata and controls

102 lines (72 loc) · 2.5 KB
Raw

Quick Start (Your Setup)

Your LiteLLM is already configured with AWS Bedrock models! 🎉

Available Models

Based on your ~/.litellm/config.yaml:

  • Claude Opus 4.5 - Most capable, best for complex reasoning
  • Claude Sonnet 4.5 ⭐ (default) - Great balance of speed and capability
  • Claude Haiku 4.5 - Fastest, good for simpler tasks
  • Claude Opus 4.6 - Latest Opus
  • Claude Sonnet 4.6 - Latest Sonnet

Ready to Go!

Your .env.local is already configured to use:

  • LiteLLM URL: http://localhost:4000/v1/chat/completions
  • Model: Claude Sonnet 4.5
  • API Key: sk-1234

Steps to Run

  1. Make sure your LiteLLM proxy is running:

    # Check if it's running
curl http://localhost:4000/health

# If not, start it:
litellm --config ~/.litellm/config.yaml

  2. Test the connection:

    npm run test:litellm

    You should see "✅ Success! LiteLLM is working."

  3. Start the app:

    npm run dev

  4. Open your browser: Navigate to http://localhost:3000

Changing Models

To use a different model, edit .env.local:

# For fastest responses:
LITELLM_MODEL=Claude Haiku 4.5

# For best quality:
LITELLM_MODEL=Claude Opus 4.5

# For latest Sonnet:
LITELLM_MODEL=Claude Sonnet 4.6

Then restart the dev server.

Troubleshooting

"LiteLLM API key not configured"

  • Check that .env.local exists and has LITELLM_API_KEY=sk-1234

"Failed to call LiteLLM API"

  • Verify LiteLLM is running: curl http://localhost:4000/health
  • Check your AWS credentials are configured
  • Look at LiteLLM logs for AWS/Bedrock errors

Port 3000 already in use

PORT=3001 npm run dev

Model name not found

  • Use the exact model name from your config: "Claude Sonnet 4.5" (with spaces and capital letters)
  • Check available models: curl http://localhost:4000/models

Cost Optimization

Interview mode (2 API calls):

  • Opening greeting: ~100 tokens
  • Each message exchange: ~500-1000 tokens
  • Typical 40-min interview: ~15-20 exchanges = ~15k-20k tokens

Evaluation (1 API call):

  • Full transcript + workspace: ~5k-10k tokens

Recommendations:

  • Practice runs: Use Claude Haiku 4.5 (~$0.20 per interview)
  • Serious prep: Use Claude Sonnet 4.5 (~$3 per interview)
  • Final polish: Use Claude Opus 4.5 (~$15 per interview)

You can switch models mid-session by editing .env.local and restarting.

Ready? Run npm run dev and open http://localhost:3000! 🚀