Kilo Code
Also known as: Kilo, Kilocode, kilocode.ai
Open coding agent with orchestrator sub-agents, browser/terminal control, and an --auto CI mode.
Kilo Code is an open source AI coding agent that runs across VS Code, JetBrains IDEs (IntelliJ, PyCharm, WebStorm), the command line, the cloud, and Slack. It takes instructions in plain language and carries out development tasks across several steps, including creating and editing files, running terminal commands, and testing in a browser. The project was forked from Cline and Roo Code and is now developed independently by Kilo-Org.
Its headline feature is Orchestrator mode. Instead of one agent doing everything in sequence, Orchestrator breaks a complex task into subtasks and routes each one to a specialist mode: Architect for planning, Code for implementation, and Debug for diagnosis. Subagents can run in parallel across separate git worktrees with inline diff review, and you can approve each step or turn on auto approval for trusted workflows.
Model choice is central to the pitch. Kilo connects to more than 500 models through OpenRouter and direct providers, including Claude Opus 4.7, Claude Sonnet 4.6, GPT-5.5, and Gemini 3.1 Pro, all at provider rates with zero markup. You can switch models in the middle of a task, and an Auto Model feature routes each task to a suitable model.
The platform reaches well beyond code generation. A Memory Bank stores architectural decisions and project conventions in repository markdown so the agent can rebuild context across sessions. Cloud Agents handle headless background execution, a dedicated Security Agent runs reachability analysis to prioritize exploitable vulnerabilities, and a Code Reviewer automates pull request review. Support for MCP servers adds custom tools.
Kilo Code was started by Sid Sijbrandij, a co-founder and former CEO of GitLab, and Scott Breitenother. It raised an $8 million seed round in December 2025 led by Cota Capital, with General Catalyst, Breakers, and Quiet Capital participating. The agent is used by millions of developers and ranks first on OpenRouter by coding token volume, and reported adopters include engineering teams at Meta, Amazon, Airbnb, PayPal, and Red Hat. The extension is free and open source. AI inference is billed separately, either by bringing your own provider key at cost or through Kilo Pass managed credits that start at $19 per month.
Vendor details
Canonical URL
https://kilo.ai
Category
Coding agent
Subcategory
Open source AI coding agent / agentic engineering platform
Funding status
$8M seed round (December 2025) led by Cota Capital, with General Catalyst, Breakers, and Quiet Capital. Cofounded by Sid Sijbrandij (GitLab cofounder and former CEO) and Scott Breitenother. 23.7k+ GitHub stars, used by millions of developers, and ranked #1 on OpenRouter by coding token volume.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Runs as a VS Code extension, a JetBrains plugin (IntelliJ, PyCharm, WebStorm), a CLI, in the cloud, and via Slack. Connects to 500+ models through OpenRouter and direct providers (Anthropic Claude, OpenAI, Google Gemini) using bring your own key at provider rates with zero markup, switchable in the middle of a task. Supports MCP servers for custom tools and GitHub for automated code review. Kilo Pass offers managed inference credits, and KiloClaw provides managed OpenClaw hosting as a separate product.
In practice
You're mid-feature and the task really needs planning, building, and debugging handled separately. Kilo's Orchestrator splits it into subtasks and routes each to a specialist mode instead of one agent fumbling through all three.
Your team is boxed into one model's pricing and limits. Kilo connects to more than 500 models at provider rates, and you can switch models in the middle of a task when one stalls.
The agent forgets your architecture every new session. Kilo's Memory Bank stores your decisions and conventions in repository markdown, so it rebuilds context instead of starting from scratch each time.
Sources & related URLs
Agentic Index coverage score
9.5 / 14 capabilities · 68%
| Integrations & Tool CallingRuns as an extension in Visual Studio Code and JetBrains and as a standalone command line tool, controls the terminal and the browser to run commands and automate the web, connects to chat channels like Slack, Telegram, and Discord through its cloud agent, and installs tools from a Model Context Protocol marketplace, broad named integrations with real action. | Full |
|---|---|
| Workflow OrchestrationBreaks a complex task into subtasks and delegates them to specialist agents for planning, implementation, and debugging, runs parallel tool calls and subagents through an Agent Manager, and can run fully autonomously in continuous integration pipelines with an auto flag, genuine autonomous multi agent orchestration. | Full |
| Knowledge Grounding & RAGMarkets an agent that understands the entire codebase and reads project context, and a Memory Bank reduces context loss between sessions, but deep codebase indexing was temporarily removed during the 2026 rebuild and is being rebuilt, so codebase grounding is present but not a headline deep index. | Partial |
| Human Oversight & GuardrailsUses human in the loop approval for file changes and terminal commands with diff based review, and its code review inspects edits file by file with line level comments, real oversight through propose and approve and review rather than a runtime guardrail enforcement engine. | Partial |
| Security, Identity & GovernanceRuns open source and fully self hosted with bring your own key so code and model calls stay under the team's control, and recent work added tighter security, usage and spend reporting, and clearer controls over agent use, real data control and emerging governance without a published certification or single sign on matrix. | Partial |
| Observability & AuditabilityProvides usage and spend reporting, model rankings, and in editor visibility into what the agent is doing, with recent focus on better usage and spend reporting and clearer controls, real visibility into agent activity and cost short of comprehensive execution tracing and export. | Partial |
| Memory & State PersistenceShips a Memory Bank that reduces context loss between sessions and supports cross platform sessions so a task can start on a phone through Slack, continue in the editor, and finish in the terminal, an explicit state persistence feature, though the Memory Bank is a legacy component pinned from an earlier version. | Partial |
| Deployment & Data ResidencyIs open source under the Apache license for the extension and the MIT license for the command line tool, runs fully locally with support for local models through Ollama and bring your own key, and ships self contained binaries including static builds for minimal containers, real self hosting and data residency. | Full |
| Prebuilt Agents, Templates & PacksShips specialized agents that you switch between for coding, planning, asking, and debugging and lets teams define custom agents in markdown files, a prebuilt agent roster with custom scaffolding, though the browsable marketplace it offers is for Model Context Protocol servers rather than agents. | Partial |
| Triggers & Channel CoverageReaches developers across Visual Studio Code, JetBrains, a command line, a hosted cloud agent, chat channels like Slack, Telegram, and Discord, and voice, and triggers automated code review on pull requests and autonomous runs in continuous integration pipelines, broad multi channel coverage with event driven triggers. | Full |
| Model Flexibility & RoutingRoutes across more than five hundred models from over sixty providers through the Kilo Gateway at zero markup, supports switching models mid task, and uses an Auto Model router that classifies each session and picks an efficient, frontier, balanced, or free model, a genuine multi provider routing gateway. | Full |
| APIs, SDKs & MCP ExtensibilityIs open source with a first class Model Context Protocol marketplace, a command line tool distributed on package managers, and custom agents defined in markdown, but the Kilo Gateway is scoped to the editor and command line rather than a general public build on application programming interface, so extensibility is strong but not a full developer platform. | Partial |
| Testing, Debugging & OptimizationShips a Debug agent that reads errors, traces issues, and suggests fixes, self checks by reviewing and correcting its own work, and runs automated code review and security analysis on pull requests, real debugging and review capability rather than a dedicated test generation and quality engine. | Partial |
| Browser & Computer UseControls the terminal to run shell commands and controls the browser to test and automate the web, including through its always on cloud agent, real code execution and browser control, though browser automation is a supporting capability rather than the core product. | Partial |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
Kilo Code disclosed a security incident involving its business intelligence provider, Metabase, which impacted Kilo Code customer data. The company was notified of the breach on August 6, 2026, and subsequently alerted its users.
Bears on: Security / enterprise
View sourceKilo Code released version 7.0.7 of its JetBrains plugin to add native support for custom OpenAI-compatible model providers. The release also improves setup validation, error dialog handling, and provider disconnection workflows.
Bears on: Integrations
View sourceKilo Code added network restriction enforcement for sandboxed agent commands running on Linux environments. The update blocks unauthorized network access across TCP, UDP, IPv4, IPv6, and descendant processes, while giving administrators the ability to whitelist specific external network destinations.
Bears on: Security / enterprise
View sourcePricing
Free / OSS extension · BYOK at zero markup · KiloClaw cloud agent ~$55/mo · Enterprise custom
Free and open source to install and use, with model usage billed by consumption. You bring your own key and pay the model provider directly at zero markup, or route through the Kilo Gateway and pay published provider rates, and the optional KiloClaw always on cloud agent is a flat monthly subscription with inference billed separately.
Included quota
The free and open source extension and command line tool include the full agent feature set, all agent modes, the Model Context Protocol marketplace, more than five hundred models, and bring your own key at zero markup. New users get 20 dollars in starter credits. Kilo Pass provides prepaid gateway credits that roll over. The optional KiloClaw cloud agent adds always on hosted execution and chat channel connectivity for a monthly fee with inference billed separately. Enterprise adds usage and spend reporting, tighter security, and centralized billing.
What is public
Public: the free and open source core, the zero markup bring your own key model, the 20 dollar starter credits, the roll over behavior of Kilo Pass credits, and the approximate KiloClaw monthly price. Not fully public: exact KiloClaw tiers, credit denominations, and enterprise pricing.
Billing mechanics
A free and open source core billed only by model usage, where you bring your own key at zero markup or use prepaid gateway credits that roll over, plus an optional flat monthly cloud agent subscription with inference billed separately, and custom enterprise pricing.
Cost watchouts
Kilo itself is free, but model usage is not, so cost tracks token consumption, and the orchestrator and subagents can multiply token use since a task that costs fifty cents with a single agent can cost more than a dollar and a half through delegation. The KiloClaw cloud agent adds a monthly subscription plus separately billed inference. Kilo Pass credits are tied to the Kilo Gateway for the editor and command line, not a general provider key.
Variable cost rationale
The software is free, so essentially all cost is model usage billed by consumption, and it scales directly with how much the agents run and which models they call. Orchestration and subagents multiply token use, and the cloud agent bills inference separately. Bring your own key routes spend to the provider rather than removing it, so exposure to usage driven cost is high.
Additional watchouts
Model token cost is the real expense and scales with usage and orchestration depth. KiloClaw adds a monthly fee plus separate inference. Kilo Pass credits work only through the Kilo Gateway for the editor and command line.
Overage / add-ons
There is no overage in the subscription sense for the core tool. When prepaid Kilo Pass credits run low, users top them up or bring their own provider key and pay the provider directly, and credits roll over rather than expiring.
Sales call required
No, self serve available
Free / trial
The extension and command line tool are free and open source to install and use. New users receive 20 dollars in starter credits, and bringing your own model key incurs no Kilo fee beyond the provider's own token cost.
Lowest paid plan
There is no required paid plan for the core tool. The lowest recurring paid product is the optional KiloClaw always on cloud agent at about 55 dollars a month, above a free and open source core.
Commercial notes
Kilo's open core and zero markup gateway make it one of the most cost transparent options in the lane, with spend tracking model usage directly rather than a seat fee. The tradeoff is variable and sometimes unpredictable cost, since orchestration and subagents consume more tokens, so teams need to watch balances on complex workflows.
Key ambiguities
Because the tool is free but models are paid by usage, the real monthly cost depends entirely on how many tokens the agents consume and which models they call, which the orchestrator can increase. KiloClaw pricing and enterprise pricing are less fully documented.
Cancellation / refund
The core extension and command line tool are free and open source with nothing to cancel. Kilo Pass credits roll over rather than expiring monthly. The KiloClaw cloud subscription has no contract and a one day free trial, so it can be cancelled by the user.
Missing data
Exact KiloClaw tiers, Kilo Pass credit denominations, and enterprise pricing are not fully published and depend on usage.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose team of ten specialized AI agents…
- Ageiro — Early stage enterprise platform that turns business goals into…
- Aider — Open-source, model-agnostic terminal coding agent that works…
- Anthropic Claude Code — Anthropic's terminal and IDE-native coding agent with sub-agent…
- AppFactor — Agentic orchestration platform that autonomously maintains and…
Alternatives to Kilo Code
The closest documented capability profiles to Kilo Code among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Zed9.5 / 14Fuller documented coverage on APIs, SDKs & MCP Extensibility
- Cosine9.0 / 14Fuller documented coverage on Knowledge Grounding & RAG
- Goose9.0 / 14Fuller documented coverage on APIs, SDKs & MCP Extensibility
- Tembo10.0 / 14Fuller documented coverage on Knowledge Grounding & RAG and Human Oversight & Guardrails
- OpenHands11.5 / 14Fuller documented coverage on Security, Identity & Governance and Observability & Auditability
- Potpie AI9.5 / 14Fuller documented coverage on Knowledge Grounding & RAG and Testing, Debugging & Optimization
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded