Drops in transparently
Hooks into Claude Code (and other agents). The model continues to call its usual tools — TokenShift silently rewrites the response so each call costs far fewer tokens, with no change to how the agent is used.
TokenShift
AI coding agents waste tokens reading git diff, test output, and build logs. TokenShift sits between the agent and the shell, so the model only sees what matters.
Nothing is lost
The full output is cached locally and addressable. If the model ever needs the original, it asks for it by name. The compressed view is a summary, not a hard truncation.
Encrypted on the wire
Every telemetry record is hybrid-encrypted client-side against a public key pinned in the binary; only PointFive’s KMS holds the private key. Fail-closed allowlist on every field. See Security.
MDM-friendly
Distributed through Homebrew, apt, yum, MSI. Bound to a tenant via a four-field enrollment manifest your MDM pushes. No control plane to manage. See MDM rollout.
Where to next
Section titled “Where to next”Install in 60 seconds
Drop TokenShift into Claude Code on a single machine and see it work. Quickstart →
What it does
The architecture in one page — proxy, hook, recovery, telemetry. How it works →
For your security team
What goes on the wire, how it’s encrypted, and what stays on the device. Security overview →