mo2g/token-insight repository overview

⁠Token Insight

Token Insight is a local-first token usage analytics stack for AI coding tools.

It scans local artifacts, normalizes usage events, stores them in SQLite, and serves a polished dashboard with deep filtering, trend analysis, source breakdowns, and social share cards.

⁠Why use Token Insight

• Local-first by default, with no mandatory cloud upload
• Multi-source ingestion across AI coding tools
• Clean dashboard for trends, model usage, sources, projects, and cost estimates
• SQLite-backed storage for easy inspection and portability
• Docker-friendly deployment for quick local setup

⁠Supported sources

Token Insight can aggregate usage artifacts from tools such as:

• Codex
• Claude
• Gemini
• Cursor
• LiteLLM

More adapters can be added over time.

⁠Quick start

Run the container:

docker run --rm -p 8787:8787 \
  -v $HOME:/host-home:ro \
  -e TOKEN_INSIGHT_HOME=/host-home \
  -e TOKEN_INSIGHT_SOURCE_ROOT_LITELLM=/host-home/.litellm \
  mo2g/token-insight:latest

Then open:

http://localhost:8787

⁠Docker Compose

docker compose up --build

Default port:

8787

⁠Important paths and environment variables

The container mounts your home directory as read-only at:

/host-home

Common environment variables:

• TOKEN_INSIGHT_HOME=/host-home
• TOKEN_INSIGHT_SOURCE_ROOT_LITELLM=/host-home/.litellm
• TOKEN_INSIGHT_DATA_DIR=/var/lib/token-insight/data
• TOKEN_INSIGHT_CACHE_DIR=/var/lib/token-insight/cache

Override source roots if your local artifacts live elsewhere.

⁠What you get

• Usage trends over time
• Model and provider breakdowns
• Source health and ingestion status
• Project-level analysis
• Exportable normalized datasets
• Shareable social images

⁠Image tags

Recommended tags:

• latest
• versioned tags such as 0.1.0

For production-style usage, prefer a versioned tag.

⁠Project links

• GitHub: https://github.com/mo2g/token-insight⁠
• Issues and feature requests: https://github.com/mo2g/token-insight/issues⁠

⁠Notes

• The app runs locally by default
• Missing source paths are skipped automatically
• Cost estimation uses built-in pricing data and can refresh remote pricing when network is available