- What do IDE, CLI, BYOK, MCP, and friends mean?
- IDE = integrated development environment (the editor app where you write code — Cursor, VS Code, etc.). CLI = command-line interface (typing commands in a terminal). API = application programming interface (a documented way for one program to call another over the network — e.g. your agent asking a model provider for completions). BYOK = bring your own key (you pay a model provider directly and plug the API key into a tool). OSS = open-source software (code you can inspect, modify, and usually run yourself — often free as in price, but the important part is the license and transparency). MCP = Model Context Protocol (a standard way for agents to call external tools and data). Inference = the API call that runs a model and returns tokens. GPU cloud = rented graphics hardware that hosts those models. FAQ itself just means frequently asked questions — start here anytime a card feels like alphabet soup.
- What is vibe coding?
- Vibe coding is building software by describing intent in natural language and letting AI generate and iterate on the code — often accepting changes with light review. Andrej Karpathy popularized the term (Feb 2025) for that accept-all workflow in tools like Cursor. In this directory, “vibe” shelves lean toward browser-based or prompt-to-ship environments — such as v0, Lovable, or Anything — that abstract infrastructure and deploy. Those builders trade granular architectural control for rapid prototyping; local agentic IDEs sit on the other end of the same spectrum.
- What is the difference between vibe coding and agentic coding?
- Vibe coding thrives in hosted, abstracted cloud environments for rapid app generation. Agentic coding deploys autonomous systems inside your local infrastructure — integrating with Git, language servers, and the file system. Tools like Cursor, Devin Desktop, Kilo Code, and Aider often pair with orchestrators that spawn sub-agents in isolated Git worktrees to plan, execute, and test before pull request review.
- Which AI coding agent is best in 2026?
- Depends on your workflow — my personal top three: (1) Cursor is the daily driver — Composer, Design Mode, IDE and cloud agents, CLI/SDK, and an iOS companion to kick off runs on the go. (2) Kilo Code is second, mainly for Auto Free model routing — the golden feature for pet projects, hobby builds, or scaffolding something serious without burning paid keys. (3) Cline is third; ClinePass is strong value and includes the SDK, so you can run good cheap models wherever you are, not only inside the VS Code plugin. Claude Code remains the terminal benchmark for top-tier reasoning and parallel subagents. Devin Desktop is a strong pick for a unified agent command center. Enterprise teams may lean toward Zencoder or Google Antigravity for compliance and multi-model review.
- What are AGENTS.md, SKILL.md, and MCP?
- AGENTS.md is project-root markdown for high-level architectural guidelines and behavioral guardrails. SKILL.md files are modular, executable procedural instructions agents load on demand to save token context. The Model Context Protocol (MCP) is an open standard for agents to interface with external tools, APIs, and data sources without bespoke integration scripts. Tool cards on this page note AGENTS.md and SKILL.md support per product.
- What are the best free or open-source AI coding agents?
- Kilo Code’s Auto Free router is my favorite zero-cost path for hobby and scaffolding work; Kilo Pass adds predictable credit plans with bonus credits when you outgrow free routing. Aider, Goose, Opencode, and OpenHands are solid BYOK options too. Cursor includes a basic free tier if you want a polished IDE without bringing your own keys. Cline stays free as a VS Code plugin (BYOK), and ClinePass adds a great-value subscription with SDK access for cheap models outside the editor. Emdash and Orca are free MIT-licensed desktop ADEs for parallel agents in isolated worktrees. Dropstone’s CLI runtime routes to evaluated open-weight models with flat pricing. Each card lists freeOptions — tiers change often.
- What is a coding agent orchestrator?
- An orchestrator is a control plane for running multiple autonomous agents at once. Long-horizon tasks in one terminal can cause branch conflicts and race conditions. Orchestrators like Emdash, Superset, Maestro, and Zencoder provision isolated environments — often dedicated Git worktrees — aggregate outputs, and give you one dashboard to review diffs and approve pull requests across parallel runs.