Kilo Code
FreemiumCode with AI without breaking the bank.
Published 27 September 2026
Scores
Popularity3/5
26K+ GitHub stars is solid traction for a product that only reached general availability in April 2026, but it remains well behind the scale of OpenCode, Cline, and GitHub Copilot in this subcategory.
Learning Curve3/5
The core extension installs like any IDE plugin, but the number of agent modes, an Orchestrator layer, and a multi-tier pricing surface (platform plan + inference plan + optional cloud compute) add more up-front decisions than a single-mode competitor.
Flexibility5/5
500+ models across 60+ providers with mid-task switching, visible prompts, BYOK/local-model support, and multiple specialized agent modes give it very open-ended configurability.
Performance4/5
Specialized modes and the Orchestrator's task-splitting approach handle structured, multi-step work well, though output quality still ultimately depends on whichever underlying model is selected for a given task.
Portability4/5
MIT-licensed, model-agnostic, and available across VS Code, JetBrains, CLI, and cloud with no required vendor lock-in, though the cloud agent/reviewer add-ons are Kilo-hosted services rather than self-hostable.
About Kilo Code
Kilo Code is an open-source (MIT-licensed) AI coding agent that reached general availability in April 2026, built on top of the OpenCode foundation. It ships as extensions for VS Code and JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm), a standalone CLI, and cloud agent infrastructure, positioning itself as an "agentic engineering platform" rather than a single-surface tool.
The agent connects to 500+ models across 60+ providers with mid-task model switching, and prices inference at exact provider rates with zero markup, whether paid through the optional Kilo Pass subscription, pay-as-you-go via the Kilo Gateway, bring-your-own-key, or free local models through Ollama. Specialized agent modes cover Code (production-ready writing and refactoring), Architect (complex feature planning), Debug, Ask, and Custom modes, with an Orchestrator mode that breaks larger tasks into coordinated subtasks across planner, coder, and debugger subagents. Prompts and context windows are visible rather than hidden, and the platform extends into cloud agents, a code reviewer product, Slack integration, and pay-per-second cloud compute for autonomous CI/CD-style runs.
In August 2026, a critical zero-day (CVE-2026-72898) in Metabase, a third-party analytics vendor used internally rather than a Kilo Code product vulnerability, exposed names and emails for a subset of Kilo Code users and some Slack tokens, which were invalidated after discovery. The underlying flaw affected many organizations using Metabase and was not a failure in Kilo Code's own codebase or agent security model.
Key Features
- 500+ models across 60+ providers, zero markup on inference, mid-task model switching
- Specialized agent modes: Code, Architect, Debug, Ask, Custom
- Orchestrator mode splits complex tasks across coordinated planner/coder/debugger subagents
- Available as VS Code and JetBrains extensions, a standalone CLI, and cloud agents
- Visible prompts and context windows — no hidden system prompt
- Built-in AI code reviewer product and Slack integration
- Pay-per-second cloud compute for autonomous CI/CD-style agent runs
- Free local model support via Ollama alongside BYOK and hosted options
Pros
- Zero-markup inference pricing across every paid path (Kilo Pass, Gateway, or BYOK) is unusually transparent for the category
- Broadest surface coverage among the open-source, non-vendor coding agents — IDEs, CLI, and cloud in one product
- Specialized agent modes and an Orchestrator mode give more structured task-breakdown than a single chat-style agent
- Fully open source (MIT) with visible prompts, appealing to teams wanting auditability
- Fast-growing community (26K+ stars) for a product that only reached GA in April 2026
Cons
- The August 2026 Metabase third-party breach, while not Kilo's own security failure, exposed some user names/emails and Slack tokens and is worth users being aware of
- Pricing surface is more complex than most competitors — separate platform plans, inference plans, and cloud-compute rates to understand
- Younger product (GA April 2026) than most other entries in the subcategory, with less of a track record
- Cloud agent and code-reviewer add-ons introduce per-hour billing on top of the core subscription/BYOK choice
Kilo Code Pricing
Freemium- · Open-source VS Code, JetBrains, and CLI extensions
- · Bring your own API keys or run free/local models via Ollama
- · Usage billed separately at provider rates with no markup
- · Per-user pricing on top of an inference plan
- · Team collaboration, shared agent modes, usage analytics, centralized billing
- · Up to $26.60 in monthly inference credits
- · Zero markup on 500+ models across 60+ providers
- · Up to $68.60 in monthly inference credits
- · Up to $278.60 in monthly inference credits
- · SSO/OIDC/SCIM, audit logs, model/provider restrictions, SLA and dedicated support
Tech Stacks with Kilo Code
VS Code AI Dev Stack
DeveloperVS Code as the editor, with your choice of AI extension (GitHub Copilot by default, or Cline, Kilo Code, Claude Code, or Codex) and your pick of GitHub or GitLab for version control.
Development
Tools Related to Kilo Code
Works well with Kilo Code(1)
Kilo Code uses OpenRouter, a featured integration in OpenRouter's own materials, to route model calls across providers.
Alternatives to Kilo Code(12)
Kilo Code is model-agnostic across VS Code, JetBrains, CLI, and cloud; Qwen Code is Alibaba's terminal agent optimized for Qwen3-Coder. Kilo Code for editor integration and model choice, Qwen Code for a Qwen-tuned CLI.
Kilo Code is a full-featured agent with specialized modes across editors and cloud; Pi is a minimal terminal harness with four core tools and a tiny system prompt. Kilo Code works out of the box, Pi is a base to build on.
Kilo Code spans VS Code, JetBrains, CLI, and cloud with 500+ models at zero markup; oh-my-pi is a terminal-only Pi fork with LSP/DAP and browser automation. Kilo Code for editor users, oh-my-pi for terminal power users.
Grok Build is SpaceXAI's terminal agent with up to eight parallel worktree sub-agents; Kilo Code is an MIT-licensed agent across VS Code, JetBrains, CLI, and cloud with 500+ models. Grok Build for parallel CLI work, Kilo Code for editor coverage.
DeepSeek's harness is a plugin-built terminal agent centred on DeepSeek models; Kilo Code spans VS Code, JetBrains, CLI, and cloud with 500+ models at zero markup. Kilo Code offers broader editors and models, DeepSeek's harness a lean vendor CLI.
Kilo Code is an MIT-licensed agent with 500+ models across editors and the CLI; ZCode is Z.ai's free desktop agent built around GLM. Kilo Code wins on model and editor choice, ZCode on a free bundled model.
Learning Resources
No resources yet — check back soon.