FAQ

What is crewkit?

crewkit is the institutional memory and governance layer for AI-assisted engineering: agents work as teammates, prior work is one search away, and every session is attributed to the exact agent version that ran it. Teams using Claude Code get shared project context, enforced conventions, and a record of what each agent version actually did.

How does crewkit work with Claude Code?

crewkit wraps Claude Code with a CLI and TUI. When you run crewkit code, it syncs your team's agent configurations (agents, skills, rules, commands) from the crewkit platform, launches Claude Code with those configs, and tracks session data (tokens, costs, tool usage, summaries) for your dashboard.

Is crewkit open source?

The CLI is proprietary and distributed as prebuilt binaries. The public karibew/crewkit-cli repository hosts releases and the issue tracker. The dashboard and API are hosted services included with your crewkit plan.

What does crewkit cost?

Free for individual developers. Team plans are per-seat — see crewkit.io/pricing for current plans.

What data does crewkit collect from my sessions?

crewkit collects session metadata: token counts, costs, model usage, tool invocations, durations, and AI-generated summaries. Conversation content (your prompts and responses) can optionally be stored as transcripts for review. Organizations can configure privacy settings to control what data is collected and retained. See our privacy policy for current data retention details.

Can I use crewkit without the dashboard?

Yes. The CLI works standalone for agent sync and session tracking. The dashboard provides analytics, team management, and experiment features on top of the CLI data.

How does 3-tier resource inheritance work?

Resources (agents, skills, rules) can exist at three levels: Platform (crewkit-managed base), Organization (team customs), and Project (project-specific). Each level can inherit from the one above, adding its own content. This lets you set company-wide standards while allowing project-specific customization. See the inheritance guide for details.

What is coaching mode?

Coaching mode is a role-based configuration that changes how AI agents interact with junior developers. Instead of writing code directly, agents in coaching mode guide developers step-by-step, ask clarifying questions, and explain concepts. It is automatically applied based on the developer's role in the organization.

How do A/B experiments work?

Create an experiment for any resource (agent, skill, etc.). crewkit splits sessions between the current version (control) and a modified version (variant). Statistical analysis tracks cost, tokens, and quality scores, with p-values for significance. When results are significant, you can deploy the winning variant. See the experiments guide for a full walkthrough.

Does crewkit work with monorepos and multi-repo projects?

Yes. crewkit supports both single-repo projects and workspace projects that span multiple repositories. The CLI auto-detects project structure from git remotes and can initialize workspaces for multi-repo setups. See the workspaces guide for more.

What platforms does crewkit support?

The CLI runs on macOS (Intel and Apple Silicon), Linux (x64), and Windows (x64, via WSL or native). Install via Homebrew, npm, Chocolatey, or the install script. See the installation guide for platform-specific instructions.

How do playbooks work?

Playbooks are collections of conventions (coding standards, patterns, and rules) that can be subscribed to at the organization level and assigned to projects. They enforce team standards automatically. Conventions can be extracted from existing code using AI analysis. See the playbooks guide for details.

Is my code sent to crewkit servers?

By default, crewkit sends session metadata — costs, tokens, tool usage, and outcomes — not your repository. Full conversation transcripts are captured only when your organization enables it, and a transcript can include code that was read, written, or discussed during the session. Analyzed conversations are indexed server-side so your team can search prior work; sessions marked sensitive or private are excluded from that index. The CLI itself runs locally alongside Claude Code and never uploads your working tree.

Can I import existing Claude Code sessions?

Yes. The crewkit sessions import command discovers and imports historical JSONL transcripts from your local Claude Code history, giving you retroactive analytics.

How do I get started?

Install the CLI with brew install karibew/tap/crewkit, then run crewkit code in any project. The CLI handles authentication, project detection, and agent sync automatically. See the quickstart guide for a full walkthrough.