tokenwatch-docs

FAQ

Does TokenWatch send my data anywhere?

No. TokenWatch runs entirely on your machine. It reads local files, calculates costs locally, and serves a dashboard only accessible at localhost:7842. Nothing is sent to any server.

The only external requests it makes are:

• Fetching live model prices from LiteLLM’s public GitHub file (no personal data)
• Checking PyPI for version updates (no personal data)
• License validation when you enter a Pro key (only the key is sent)

How does TokenWatch calculate costs?

It reads the usage field from Claude Code’s JSONL session logs. Each assistant response contains:

• input_tokens
• output_tokens
• cache_read_input_tokens
• cache_creation_input_tokens

These are multiplied by Anthropic’s current per-token prices (fetched automatically).

Does it work with Cursor, Windsurf, or other AI editors?

Not yet. TokenWatch currently reads Claude Code’s specific log format. Support for other tools is planned for a future version.

What happens if Anthropic changes their pricing?

TokenWatch fetches live prices from LiteLLM’s community-maintained price list automatically on startup. When Anthropic changes prices, LiteLLM usually updates within hours. You can force a refresh by clicking ↺ Refresh in the prices panel.

Can I use TokenWatch on multiple machines?

Yes. Install it on each machine separately. Each instance reads that machine’s local Claude Code logs independently.

How do I move my config to a new machine?

Copy ~/.claude/tokenwatch_config.json to the same path on the new machine.

My cost numbers seem too high / too low

Click ↺ Refresh in the prices panel to make sure you’re using current prices. Then visit http://localhost:7842/api/debug to verify the log format is being read correctly.

If numbers still seem wrong, email info@consulting-ai.tech with the output of the debug endpoint.

Can I use this for a team?

The Pro plan includes team cost aggregation (coming soon — currently in development). For now, each team member installs individually.

Is there a refund policy?

Yes — if TokenWatch doesn’t work for you, email info@consulting-ai.tech within 7 days of payment for a full refund, no questions asked.