Tool
Kilo Code
An open-source coding agent that meets you in VS Code, JetBrains, or the CLI — multi-file edits, specialized agents, MCP extensions, and a long model menu, with MIT on the software you install.
What it is
Kilo Code lives in Kilo-Org/kilocode. Per the current README, it is an AI coding agent for building with AI in the editor or terminal: code generation across files, inline autocomplete, self-checking, terminal and browser control, and an MCP marketplace. Modeled agents include Code, Plan, Ask, Debug, and Review (plus custom agents). The product pitch emphasizes open pricing — pick among many models, switch mid-task, and pay provider rates. There are also hosted paths (cloud agent, PR reviews, always-on “KiloClaw”) that are a different trust and billing boundary than the local extension/CLI. The CLI lineage notes a fork of OpenCode enhanced for the Kilo platform.
What it replaces
It replaces some proprietary editor lock-in when you want an MIT agent in VS Code /
JetBrains / CLI that can act on a repo — competing in the same space as
Cline, Goose,
Aider, and the agent modes inside
Cursor. It does not replace code review, CI, or
human judgment when the agent proposes a destructive command — especially
kilo run --auto, which the docs warn disables permission prompts.
License
The Kilo-Org/kilocode repository is licensed under the MIT License (verified from the project LICENSE; copyright notices include Kilo Code and opencode lineage). Model providers, marketplace MCP servers, and any Kilo cloud surfaces still have their own terms and bills.
Where it falls over
- Model bills are real — “zero markup” still means you pay the provider for tokens.
- Autonomy without guardrails (
--auto, broad tool grants) can edit or run the wrong thing. - Younger ecosystem vs Cursor’s polished product surface — expect sharper edges and faster-moving docs.
- Cloud agent / reviews / always-on paths are not the same trust story as offline local agent use.
- Output quality caps at the model you chose; local small models ≠ frontier reasoning.
Who it's for
Developers who want an MIT coding agent they can run in VS Code, JetBrains, or CLI, will review diffs, and will own model spend. Wrong tool if you only wanted tab-complete inside a proprietary IDE with a single vendor SLA and no interest in agent permissions.
Related in this cluster
- Cline — open coding agent in the editor
- Goose — Apache-2.0 extensible agent runtime
- Aider — terminal pair-programmer for git repos
- Continue — editor companion; acquired by Cursor
- Cursor — proprietary AI code editor to compare against
Verdict
Use Kilo Code when an MIT agent across VS Code / JetBrains / CLI beats locked-in autocomplete, and you will constrain tools, review risky actions, and watch the model bill. Skip the autonomy cosplay if you cannot sandbox or read the diff — and treat cloud add-ons as a separate decision from the open client.