Summary

Caliber addresses AI agent configuration drift and staleness in software projects by continuously syncing AI setup files with evolving codebases. It targets development teams using AI coding assistants like Claude Code, Cursor, and Copilot, ensuring accurate, up-to-date agent context for improved performance and team consistency.

How It Works

The core approach involves caliber bootstrap or caliber init analyzing project structure to generate tailored AI agent configurations. Pre-commit hooks enable continuous synchronization via caliber refresh. A deterministic scoring system (caliber score) audits configuration quality against the filesystem without LLM calls, presenting changes as diffs for review and optional chat-based refinement. Session learning captures AI coding patterns to enhance future configurations.

Quick Start & Requirements

• Primary Install/Run: npx @rely-ai/caliber bootstrap (one-time setup) or caliber init (CLI wizard).
• Prerequisites: Node.js >= 20.
• Resource Footprint: Bootstrap is ~2 seconds. Generation requires an AI subscription or API key; scoring and bootstrap are 100% local.
• Links: CONTRIBUTING.md available.

Highlighted Details

• Deterministic Scoring: caliber score audits config quality against the project filesystem without LLM calls, providing detailed, actionable feedback.
• Chat-Based Refinement: Generated configurations can be refined via natural language during the review process.
• Session Learning: Captures AI coding session events (corrections, patterns) to generate CALIBER_LEARNINGS.md for improved future configurations.
• Auto-Refresh Hooks: Git pre-commit and session-end hooks automatically update configurations based on code changes.
• Team Onboarding: Automatically prompts new team members to set up Caliber.
• Fully Reversible: Features automatic backups, score regression guard, caliber undo, and caliber uninstall for safe management.
• Broad Compatibility: Supports numerous languages (TypeScript, Python, Go, Rust, Java, Ruby, Terraform) and AI agents (Claude Code, Cursor, Codex, Copilot).

Maintenance & Community

No specific details on maintainers, sponsorships, or community channels (Discord/Slack) were found in the README. CONTRIBUTING.md is referenced.

Licensing & Compatibility

• License: MIT.
• Compatibility: The MIT license is permissive, allowing for commercial use and integration with closed-source projects.

Limitations & Caveats

Configuration generation requires an existing AI subscription or API key (Anthropic, OpenAI, Vertex AI), though bootstrap and scoring are fully local. Anonymous usage analytics are enabled by default but can be disabled. The README does not explicitly state alpha/beta status.