Kilo Code review

Kilo Code is an open-source coding agent delivered through IDE extensions and plugins, a terminal CLI, cloud agents, a desktop app and Slack. It sits between a developer and the model provider: the agent works with a codebase, while users choose a model, connect provider keys, use local models or route requests through Kilo Gateway. The homepage says Gateway supports more than 500 models and charges provider rates without inference markup.

The agent is not just a hosted chat interface. Its MIT-licensed source can be inspected, modified and self-hosted, while the hosted product adds cloud agents, centralized team features and billing options. Specialized Code, Plan and Debug agents support implementation, planning and troubleshooting workflows. Auto Model can route work through Efficient, Frontier or Free strategies. In a stack, Kilo is an agent layer for development workflows, not a replacement for the underlying model API, IDE or source-control system.

The published integration surfaces are IDE extensions and plugins, a shell CLI, browser-based cloud agents and Slack. The homepage names VS Code, Cursor, Windsurf and Open VSX editors, plus JetBrains IDEs including IntelliJ IDEA, PyCharm and WebStorm. The CLI is presented for shells, scripts and CI jobs; its install command is curl -fsSL https://kilo.ai/cli/install | bash. The homepage also lists npm, pnpm, bun, brew and paru installation options.

The homepage names no public REST API or language SDK. Treat Kilo as an agent you integrate into developer workflows rather than assuming a service SDK is available; confirm SDK support and supported protocols in the vendor's docs before building a custom integration. It advertises cloud agents, Slack mentions and a Gateway, but does not describe these as webhook or REST interfaces. Docs should also be checked for extension permissions, CLI configuration, model-provider setup and how agent sessions move between local and cloud environments.

The Free plan costs $0 per month and includes VS Code, JetBrains and CLI extensions, code completion and generation, and support for any AI model. Usage is billed separately, so free access to the agent does not mean free inference. Auto Free, BYOK or local also costs $0 per month; it includes free Gateway models where available, provider keys and Ollama or LM Studio support.

Kilo Gateway has no monthly platform fee, and the pricing lists exact provider rates with no markup. Kilo Pass subscriptions start at $19 per month for Starter, $49 for Pro and $199 for Expert. Their listed maximum credits are $26.60, $68.60 and $278.60 per month, respectively, with up to 50% bonus credits. Teams costs $15 per user per month; Enterprise pricing is custom. Credit purchases carry a 5% processing fee. Cloud compute is separate: Code Review is $0.33 per hour, Cloud Agent Docker and Small are $0.60 per hour, and Standard is $1.20 per hour. Compute is billed per second; inference is additional.

Kilo Code says its agent source is MIT-licensed, and that users can inspect, fork, modify and self-host it. That makes the agent layer easier to audit and adapt than a closed, hosted-only coding tool. The homepage also describes prompt and context visibility and says the agent does not silently switch models.

Open source does not remove every dependency. Model access still depends on a provider, a local model setup or Kilo Gateway, and hosted cloud or team features have their own service and billing relationship. The vendor does not publish a data-export procedure on the homepage. Before adopting it, ask how to export conversations, agent runs, configuration and usage records, and whether those exports remain usable outside Kilo. BYOK and local models can reduce provider dependence, but they do not by themselves establish what data the hosted surfaces retain.

Before standardizing on Kilo, confirm model-provider rate limits, retry behavior, service availability and support expectations. The homepage advertises Enterprise model and provider limits, audit logs, SSO, OIDC, SCIM and SLA commitments, but does not publish numerical rate limits or SLA terms. Teams get usage analytics and reporting, shared custom agents, centralized billing and priority support. Ask the vendor to document data handling and access controls for your code before routing sensitive repositories through hosted agents.

Reddit reports are mixed rather than conclusive. One commenter says Kilo handled a provider's rate limiting with exponential backoff. In a separate thread, a long-time user reports settings migration failures, lost configurations and unsaved API keys after an update. Those are individual accounts, not measured reliability data, but they make upgrade testing and configuration backups sensible for production rollouts. Model inference and cloud compute are billed separately, so monitor both when setting budgets.