Full compression engine
Deterministic context compression for every request. Not a limited demo — the OSS package is the full product.
Open source core
TokenPak OSS is the full product.
The core proxy, compression engine, client integrations, local dashboard, and telemetry store are all Apache 2.0 licensed. Pro adds team-scale features on top of the same core — it does not unlock the OSS engine.
Install
pip install tokenpak
tokenpak integrate claude-code --apply
Pro is delivered as the tokenpak-paid package via a separate index
(https://pypi.tokenpak.ai).
Pro is the tier name; tokenpak-paid is the package name.
What's in OSS
Local proxy, no cloud
Runs at 127.0.0.1. No data leaves your machine; no credentials stored in a third party.
Nine client integrations
Claude Code, Cursor, Cline, Continue, Aider, OpenAI SDK, Anthropic SDK, LiteLLM, Codex.
Cost tracking + vault search
Local telemetry store, A/B replay, 50 built-in compression recipes, semantic vault indexing.
OSS vs Pro
Every row canonical per the messaging standard. OSS is not a trial — it is the product.
| Feature | OSS (free) | Pro |
|---|---|---|
| Context compression (full engine) | Included | Included |
| Local proxy at 127.0.0.1 | Included | Included |
| Client integrations (9 tools) | Included | Included |
| Model routing with fallback rules | Included | Included |
| Local cost tracking + telemetry | Included | Included |
| Vault indexing + semantic search | Included | Included |
| Local dashboard | Included | Included |
| A/B testing + request replay | Included | Included |
| 50 built-in compression recipes | Included | Included |
| Team dashboard (multi-seat cost attribution) | — | Pro |
| Advanced routing policies (cost-aware, SLA-aware) | — | Pro |
| Enterprise credential mgmt (SSO, rotation hooks) | — | Pro |
| Priority support and SLA | — | Pro |
| Custom recipes + compliance rulesets | — | Pro |
Source + package
Ready to try it?
tokenpak runs entirely on your machine. Nothing to sign up for.