AI DevKit is a local operating layer for developers who run several AI coding agents at once. It does not replace those agents. It configures, supervises, and coordinates the ones you already use, so each tool shares one setup, one console, a searchable project memory, and the same set of engineering skills.
A single project config file reconciles each agent's rules, MCP servers, and skills, and an init wizard writes the project-local files you can review and commit. A CLI and a live terminal dashboard list running sessions across providers, show their detail, and send prompts or piped output (test logs, for example) to one agent or a saved group. Sessions can also be relayed through Telegram for remote control.
Project memory is a local SQLite store of decisions, conventions, and fixes, exposed to agents over MCP and the CLI so an agent searches it when a task needs context rather than carrying every rule in every prompt. The anchor skill, dev-lifecycle, walks an agent through requirements, design, planning, implementation, testing, and review, writing phase documents along the way, and companion skills add verification gates, test-first work, structured debugging, and focused review passes. More skills install from registries of other publishers.
Supported agents include Claude Code, Gemini CLI, Codex CLI, Grok Build CLI, Junie, Cline, Devin, OpenCode, Pi, Kiro CLI, Cursor, GitHub Copilot, Antigravity, Amp, and Kilo Code, with Roo Code in testing. Remote control covers a subset of them. AI DevKit runs locally with no hosted service or telemetry.
Features
- One config for every agent: a single project file reconciles rules, MCP servers, and skills across the coding agents you pick
- Agent console: a live terminal dashboard lists and inspects running sessions across providers
- Cross-agent messaging: send prompts, piped command output, or follow-up work to a running agent or a saved group and wait for the reply
- Retrieval-based memory: a local SQLite knowledge base of conventions and fixes, searched over MCP or the CLI only when a task needs it
- dev-lifecycle skill: guides an agent through requirements, design, planning, implementation, testing, and review with phase documents
- Verification gate: the verify skill blocks a completion claim without fresh test or build output
- Engineering skills: TDD, structured debugging, commit hygiene, code documentation, simplification, security review, and technical-writing passes
- Skill registries: add skills from 30+ publishers, including Anthropic, Vercel, Supabase, Microsoft, and Google
- Remote channels: relay a session through Telegram to operate an agent from a phone
