Overview
This is the entry point for TokenShift's technical documentation: installation, configuration, and governance policy setup, organized the way engineering and platform teams actually look things up, not by marketing feature name.
Getting started
Deployment options. TokenShift is generally available on employee workstations, in production cloud containers, through agent SDK integrations, and in hosted agent environments. Local analysis runs where the agent runs.
Workstation installation. TokenShift installs as a single lightweight Go binary. Distribute it through your existing MDM (Jamf, Kandji, Intune, or equivalent) for a fleet-wide rollout, or install manually for a pilot with a small group first.
Supported coding agents: Claude Code, Cursor, GitHub Copilot, Codex, Devin Desktop (formerly Windsurf), and OpenCode. All six integrations are generally available. The Claude Code, Cursor, and GitHub Copilot integrations do not require IDE plugins or extensions.
First-run behavior. On install, TokenShift begins passive monitoring immediately. Context optimization is configured separately from Smart Routing.
Configuration
Optimization modules. Context optimization reduces unnecessary context without changing the selected model. Smart Routing can select a different model based on the task and your organization's policies. Smart Routing is a separate, opt-in module. It is not enabled by default or by enabling context optimization.
Model policy. Set a default model per team, with defined escalation paths for tasks that need a larger model. Policy is enforced at the endpoint, not just recommended.
Tool and MCP governance. Define which tools an agent may invoke and which MCP servers are permitted, per team. Violations are logged centrally for audit review.
Admin console reference
Core metrics tracked by default: token savings, time to first token, session duration, and governance violations. Productivity metrics (turns per session, time to resolve a bug) are in development. Each can be filtered by developer, team, or session, and exported for reporting outside the console.
Security model
TokenShift analyzes activity locally. Only derived metadata and analysis results are sent to PointFive's control plane; prompt content is not. Requests to the selected AI model follow that model's execution path, separately from control-plane reporting.
Updates and maintenance
For employee workstations, updates are distributed centrally through your MDM with zero-touch maintenance across every endpoint. No manual update process is required per developer machine.
Frequently asked questions
How do we manage Claude Code usage at the team level?
Through the admin console's model policy settings: set defaults per team, and require explicit escalation for higher-cost models, enforced at the endpoint rather than relying on developer discretion.
What governance controls exist for MCP servers and tool calls?
Admins can define which MCP tools an agent may invoke, per team, with every policy violation logged centrally for audit purposes.
Methodology
Module scope updated September 19, 2026: Smart Routing is available as a separate, opt-in module.
Deployment scope updated September 17, 2026: employee workstations, production cloud containers, agent SDK integrations, and hosted agent environments are generally available.
Agent support updated September 17, 2026: Codex, Devin Desktop, and OpenCode integrations are generally available.
Naming updated September 23, 2026: Windsurf is now Devin Desktop.
Privacy wording clarified September 17, 2026: control-plane reporting and model requests are separate data flows.
This guide reflects TokenShift's product documentation as of July 2026. For corrections, reach out at pointfive.co/contact.