Ai Setup

This open source command line tool from Caliber generates and maintains the context files that coding agents read. Running its bootstrap command and then a setup command inside Claude Code or the Cursor CLI has the agent write CLAUDE.md, Cursor rules, AGENTS.md and Copilot instructions for the repository, show the diff, and install hooks that keep those files current. A wizard mode does the same job without an agent. It is for developers who use several agents on one codebase, including Claude Code, Cursor, Codex, OpenCode and GitHub Copilot. A pre commit hook reads each diff and refreshes the context files, and a score command audits them against the repository without calling a model, checking that paths exist, commands are real and nothing has drifted. A sync command takes a skill, rule, plugin or MCP server written for one agent and mirrors it into the others, degrading where an agent lacks a feature and never overwriting a hand edited file. A newer compaction feature, Jev from TypeSafe, scores tool calls in a session and drops stale ones instead of summarizing, and ships as a Claude Code plugin. Every write is shown as a diff first and backed up, and an undo command reverts changes. The tool is free and MIT licensed. It needs Node 20 or later and runs on an existing Claude Code or Cursor seat or on the user's own Anthropic, OpenAI, MiniMax or Vertex key, with bootstrap, scoring and sync running locally. Jev compaction requires a Vercel AI Gateway or TypeSafe key. The listed website belongs to Caliber's commercial business, which describes itself as a fractional AI board for enterprise teams and sells AI adoption measurement, a permission aware company brain, meeting capture and ROI reporting by demo. That offering has no published pricing and is separate from this free tool.