TokenPak Quickstart — First Savings in 5 Minutes¶
Get TokenPak running and see your first cost savings in under 5 minutes.
1. Install¶
pip install tokenpak
Requires Python 3.8+.
2. Start the Proxy¶
tokenpak start
Expected output:
TokenPak proxy running at http://localhost:8766
Ready to intercept API requests
The proxy is now listening for requests.
3. Point Your Client at the Proxy¶
Choose your integration method:
OpenClaw Users¶
Edit your openclaw.json:
{
"models": {
"tokenpak-anthropic/claude-sonnet-4-6": {
"provider": "anthropic",
"base_url": "http://localhost:8766",
"apiKey": "$ANTHROPIC_API_KEY"
}
}
}
Direct API Calls¶
export ANTHROPIC_BASE_URL=http://localhost:8766
Any OpenAI-Compatible Client¶
import anthropic
client = anthropic.Anthropic(
base_url="http://localhost:8766",
api_key="your-key-here"
)
4. Verify It's Working¶
tokenpak status
Check the health endpoint:
curl http://localhost:8766/health
Expected response: {"status": "healthy"}
5. See Your Savings¶
Make a few API requests through the proxy, then run:
tokenpak savings
Or check the stats endpoint:
curl http://localhost:8766/stats
What You'll See¶
- Cache hit ratio: Climbs toward 80-90% as patterns repeat
- Token compression: Compressed tokens vs. original token count
- Cost per request: Decreasing with each cached/compressed hit
- Total savings: Running tally of reduced API charges
6. Keep It Running¶
For continuous savings, keep the proxy running in the background:
nohup tokenpak start > ~/.tokenpak/proxy.log 2>&1 &
echo $! > ~/.tokenpak/proxy.pid
To stop:
kill $(cat ~/.tokenpak/proxy.pid)
Next Steps¶
- Tune cache settings: See
tokenpak configfor TTL and compression options - Monitor savings: Visit
http://localhost:8766/dashboardfor a live view - Full docs: See
GETTING-STARTED.mdfor advanced config and troubleshooting
Troubleshooting¶
"Connection refused" on http://localhost:8766
- Verify the proxy started: tokenpak status
- Check port 8766 isn't in use: lsof -i :8766
"API key invalid" errors
- Ensure ANTHROPIC_API_KEY is set: echo $ANTHROPIC_API_KEY
- Proxy is transparent — your API key must be valid
No savings showing
- Cache and compression need multiple requests to show ROI
- First request always passes through uncached
- Check tokenpak stats for live hit counts