LogoTrendingDevTools
icon of Kilo Code

Kilo Code

Open-source AI coding agent across VS Code, JetBrains, CLI, and cloud, with model choice and local inference.

Visit Website

Introduction

Overview

Kilo Code is an open-source AI coding agent that follows a developer across an IDE, terminal, and optional hosted workflows. Its official materials describe extensions for VS Code-compatible editors and JetBrains, a command-line agent for shell and CI tasks, and cloud agents that can be started from a browser. The product also has Slack and mobile surfaces for collaborating with agents. The same agent framework offers Code, Plan, Ask, and Debug modes, so a developer can choose whether the agent should work on files, propose a plan, answer questions, or investigate a problem. The agent guide describes tool access and the built-in roles; Ask is read-only, while Code and Debug can use editing and command tools.

Kilo is model-flexible rather than tied to one AI provider. The official product page advertises more than 500 models, and the pricing page separates the software platform from the costs of model inference and cloud compute. Users can choose free models where they are available, connect their own provider keys, use Kilo Gateway at provider rates with no inference markup, or run models locally. The official local-model guide supports Ollama, LM Studio, and Atomic Chat. That makes Kilo relevant both to developers who want a managed model catalog and to teams that want control over model choice, keys, or local execution. It does not make all Kilo features self-hosted: cloud agents and gateway services are hosted options, and their use is priced separately where applicable.

Key Features

  • Multiple coding surfaces: Kilo offers a VS Code extension, JetBrains plugin, CLI, optional browser-based Cloud agents, Slack integration, and a beta desktop app. The installation guide describes supported installation paths, including Open VSX for compatible editors.
  • Specialized agents: Code implements changes, Plan provides read-only planning guidance, Ask answers questions without editing files, and Debug focuses on diagnosis and repair. Teams can define custom agents with their own tool access, file permissions, and instructions (agent guide).
  • Model choice and Auto Model: Kilo supports many hosted model providers and lets users switch models during a task. Auto Model offers Free, Efficient, and Frontier strategies intended to balance model capability and cost; the selected provider and availability still affect the result (Kilo overview, pricing).
  • Local inference: The official docs support Ollama, LM Studio, and Atomic Chat for models running on the developer’s machine. The local-model page calls out privacy and offline access as benefits, while warning that hardware, setup, speed, feature support, and model quality vary (local-model guide).
  • Terminal and automation workflows: Kilo CLI can run in a shell, script, or CI job. Its official CLI page describes sandboxed auto mode, workspace-scoped writes, read-only Git protection, optional network blocking, and parallel agents. The page cautions that auto mode approves actions unless a rule blocks them, so it should be used only in trusted environments (CLI page).
  • Gateway and organization controls: Kilo Gateway provides an OpenAI-compatible endpoint, BYOK support, usage tracking, model/provider restrictions, per-user spending controls, and organization balance management (Gateway docs). The pricing page also describes team billing, usage analytics, shared agents, data privacy controls, and enterprise security features.

Use Cases

Kilo Code can help implement a feature across several files, debug a failing test, draft an architecture plan, or answer questions about a codebase. A developer can work in an IDE for interactive edits, switch to the CLI for a terminal task or CI job, or use a cloud agent when the task should run remotely. The named agents make the expected behavior clearer: Ask and Plan help reduce accidental edits during exploration, while Code and Debug are intended for implementation and troubleshooting. Custom agents can encode a team’s preferred tools and instructions.

The broad model catalog is useful when a team wants to compare models or keep an alternative provider available instead of relying on one model vendor. Bring-your-own-key support fits teams that already have provider access; Kilo Gateway can centralize multiple providers and report usage; Auto Free gives individuals a way to try models the gateway currently makes available at no credit purchase. For privacy or offline work, local models can keep code and prompts on the user’s computer when the chosen provider is genuinely local. Kilo’s own documentation notes that local models may be slower, require powerful hardware, and lack features such as prompt caching or computer use, so a local setup may not match a hosted frontier model.

For teams, the paid collaboration tier adds shared agents, centralized billing, analytics, and management controls. Enterprise plans add custom governance, SSO, audit logs, and SLA options. Cloud agents and code review provide separate remote workflows, but model inference and cloud compute are billed separately according to the current pricing page. Kilo is therefore best understood as a toolkit with different operating modes and cost components, not as one flat-fee model subscription.

Best for

Kilo Code is best for developers or teams who want one open-source coding agent across IDE and CLI workflows, flexible model providers, and the option to use local models. It is a particularly good fit for teams that need to compare inference options, centralize provider access, or run agent tasks in a sandboxed terminal. It is less suitable if the requirement is a predictable all-inclusive price for unlimited frontier-model inference, or if a team needs every cloud feature to run on its own infrastructure: Kilo’s current pricing separates inference and cloud-compute charges, and the reviewed sources do not document a self-hosted replacement for its hosted gateway or cloud-agent service.

Alternatives

Cline is a relevant alternative for developers who want an open coding agent centered on an IDE and who prefer to configure their own model provider. OpenHands is a broader open-source software-agent platform with its own local and hosted deployment choices, useful when teams are comparing autonomous engineering workflows. Gemini CLI is a terminal-based agent for users who want a command-line workflow focused on Google’s Gemini models. Kilo differs by combining IDE, JetBrains, CLI, cloud, and collaboration surfaces with multiple provider choices in one product family.

Pricing

Kilo’s platform access is Free for individuals, and the official pricing page labels the VS Code, JetBrains, and CLI agent as free and open source. Teams costs $15 per user per month; Enterprise pricing is custom. AI inference is separate: users can route to free models where available, use their own provider keys, or use Kilo Gateway at exact provider rates with no markup. The optional Kilo Pass starts at $19 per month for a monthly credit subscription; higher displayed tiers are $49 and $199 per month. Kilo Gateway itself is listed at $0 per month plus usage, and inference is billed separately from platform access.

Cloud compute is another separate charge. The current official pricing page lists code review at $0.33 per hour, Cloud Agent Docker and Small at $0.60 per hour, and Standard at $1.20 per hour; it says compute is billed per second and model inference is separate. The Auto Free route has no hosted-credit requirement but only uses free models when available. The reviewed official page does not publish a guaranteed per-user count of free requests or a fixed quota for every free model, because provider availability and limits vary (official pricing). Prices can change, so the current pricing page is the source of truth.

Frequently Asked Questions

1. Is Kilo Code free to use?

The individual platform plan is free, including the open-source VS Code, JetBrains, and CLI agent. Model usage is billed separately unless the user chooses a free model, uses a local model, or supplies provider access through BYOK. Teams, Kilo Pass, and cloud compute are optional paid components (pricing page).

2. Which coding environments does Kilo support?

The official materials list VS Code and compatible editors, JetBrains IDEs, CLI, cloud agents, Slack, and a beta desktop app. Open VSX supports compatible editors such as VSCodium, Gitpod, Eclipse Theia, and Windsurf (installation guide, homepage).

3. Can I run models locally instead of sending code to a hosted model?

Yes. Kilo’s official docs support Ollama, LM Studio, and Atomic Chat local inference. The docs warn that local models require setup and capable hardware, can be slower or less capable than larger hosted models, and may not support features such as prompt caching or computer use (local-model guide).

4. What are the main Kilo pricing components?

The platform is free for individuals; Teams is listed at $15 per user per month and Enterprise is custom. Model inference can use free models, BYOK, provider-rate Kilo Gateway, or the optional Kilo Pass starting at $19 per month. Cloud compute is billed separately at the published rates for code review and cloud-agent sizes; the pricing page describes per-second compute billing (pricing page).

5. Is Kilo’s agent safe to run in automatic mode?

Kilo CLI documents an OS sandbox that can confine writes to the workspace, protect .git as read-only, and optionally deny network access. The official CLI page says auto mode approves permission prompts unless a rule explicitly denies the action and recommends using it only in trusted environments. Review the sandbox and permission configuration before running unattended tasks (CLI documentation).

ℹ️

Affiliate Disclosure: TrendingDevTools may earn a commission if you sign up through links on this page, at no extra cost to you.

Newsletter

Join the Community

Subscribe to our newsletter for the latest news and updates