Gemini CLI Observability Hands On

한국어 README

A local development environment that sends Gemini CLI requests to a LiteLLM Proxy, which then forwards traffic to the Google Gemini API. Request/response logs can be viewed in Phoenix.

🚀 Components

• LiteLLM Proxy: OpenAI-compatible proxy that routes requests to Google Gemini.
• Phoenix: LLM observability UI.
• Postgres: Stores proxy metadata.
• gemini-cli: CLI used by developers.

📁 Key Files

• docker-compose.yml – Runs Proxy / Phoenix / Postgres
• litellm_config.yaml – Defines model aliases and actual Gemini model IDs
• .env – Google API Key + Proxy Master Key
• env.sh – Environment variables for gemini-cli

▶️ How to Run

1) Generate an API Key from Google AI Studio

• Go to https://aistudio.google.com/api-keys
• Insert the key into GEMINI_API_KEY in .env

2) Start Containers

docker compose up -d

3) Apply gemini-cli Proxy Environment

source env.sh

4) Test

gemini
# Enter /auth, type sk-1234567890, then exit
gemini --model="gemini-2.5-flash-lite" -p "hello"

🔎 Observability

Phoenix UI: 👉 http://localhost:6006

🧪 Test Calling the Proxy Directly

curl -X POST "http://localhost:4000/v1beta/models/gemini-2.5-flash-lite:generateContent" \
  -H "Authorization: Bearer $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"hi"}],"role":"user"}]}'

❗ Troubleshooting

• If the gemini-cli model name ≠ model_name, a 404/500 may occur.

• Ensure GEMINI_API_KEY in .env is a valid Google API key.

• Check LiteLLM logs:

  docker logs -f litellm