Low latency
Under 50ms compression overhead on typical agent prompts. Your agent does not feel slower.
TokenPak — local LLM proxy
Local. Cheaper. Faster.
TokenPak is a local proxy that compresses your LLM context before it hits the API — fewer tokens, lower cost, same results.
Why TokenPak
Built for the parts of an agent workload you cannot renegotiate.
One-command setup
pip install tokenpak && tokenpak setup. Interactive wizard detects your API keys, picks a compression profile, and starts the proxy. Per-client auto-integration (tokenpak integrate) is on the roadmap.
Works with what you already use
Claude Code, Cursor, Cline, Continue, Aider, OpenAI SDK, Anthropic SDK, LiteLLM, Codex. No plugin rewrites.
Local and private
No cloud component. No credentials stored. Requests still go to your model providers, but compression happens on your machine.
How it works
Three steps. No cloud, no rewrites.
Step 1
Install
pip install tokenpak. Runs at 127.0.0.1 as a local proxy.
Step 2
Setup
tokenpak setup — interactive wizard wires your keys + starts the proxy. Point your LLM client at http://127.0.0.1:8766 via one env var.
Step 3
Save
Every request is compressed deterministically; savings logged locally.
Latest release
Refreshed automatically on every release + on a daily safety-net schedule.
latest
**Retroactive SHA256SUMS attestation.**
Documentation highlights
Curated entry points sourced from tokenpak/docs.
starter
Getting Started
Install TokenPak and see savings in one command.
starter
Installation
pip install options, OS notes, troubleshooting install failures.
reference
CLI Reference
Every tokenpak verb, flag, and exit code.
reference
Architecture
Proxy-centered model; three planes; 18 subsystems at a glance.
support
Troubleshooting
Common symptoms and the fixes that work.
See savings in one command.
pip install tokenpak && tokenpak integrate claude-code --apply. No cloud, no credentials stored, no code changes.