DOCSEARCHbyAlgolia
Submit Docs

Kilo AI-powered development tool

repository·main·Indexed 42 days ago

https://github.com/kilo-org/kilo

An AI-powered development tool and agentic engineering platform. Kilo provides a GitHub Action for explaining, fixing, and reviewing issues and PRs via /kilo or /kc commands, a VS Code extension, and a gateway for AI provider integration. It includes specialized CI containers for Linux-based jobs and supports various AI models including Claude and GPT.

Tokens
359.3K
Snippets
848
Records
1.5K
Agent score
50%

What's inside Kilo

  1. Overview of the Kilo Code VS Code Extension Rebuild

    main

    The Kilo Code VS Code extension is a ground-up rebuild of the original extension, designed to use the Kilo CLI as its backend.

    Key architectural shifts in this version include:

    • Backend: Uses Kilo CLI for core logic.
    • Frontend: A Solid.js-based webview for the UI.
    • Communication: A message-based protocol between the VS Code extension host and the webview.
    • UI Components: Leverages the @kilocode/kilo-ui shared component library for consistent rendering of messages, diffs, and tools.

  2. Overview of Kilo AI Gateway features

    main

    The Kilo AI Gateway is a unified API endpoint designed to manage access to multiple AI providers. Key capabilities include:

    • Unified Access: Access models from Anthropic, OpenAI, Google, xAI, Mistral, MiniMax, and more using a single API key.
    • OpenAI Compatibility: Acts as a drop-in replacement for OpenAI's /chat/completions endpoint.
    • Streaming: Supports full Server-Sent Events (SSE) streaming with time-to-first-token tracking.
    • BYOK (Bring Your Own Key): Allows using your own provider API keys with encrypted-at-rest storage.
    • Usage & Controls: Provides per-request cost/token tracking, model allow lists, provider restrictions, and per-user daily spending limits.
    • Advanced Capabilities: Supports tool calling (with automatic repair) and Fill-in-the-middle (FIM) completions via Mistral Codestral.

  4. Available Kilo Code Platforms

    main

    Kilo Code can be integrated into various development environments and interfaces:

    • IDEs:
      • VS Code: The primary IDE integration.
      • JetBrains: Supports IntelliJ, PyCharm, WebStorm, and other JetBrains products.
    • Terminal & Cloud:
      • CLI: Terminal-based AI coding for scripts and automation.
      • Cloud Agent: Run Kilo in a cloud-hosted environment.
    • Communication & Mobile:
      • Mobile Apps: Support for iOS and Android.
      • Slack: Chat with Kilo directly within your workspace.
    • Application Development:
      • App Builder: Create full-stack applications using AI.

  5. Core Features of Kilo Code

    main

    Kilo Code includes several productivity-focused features:

    • Autocomplete: Provides inline code suggestions as you type.
    • Fast Edits: Enables quick modifications to files.
    • Code Actions: Offers AI-powered refactoring and automated fixes.
    • Task & Todo Lists: Helps break down complex development tasks into manageable steps.
    • Checkpoints: Allows you to save and restore working states during development.
    • Browser Use: Automates web-based interactions.
    • Enhance Prompt: Automatically improves your prompts for better AI responses.
    • Git Commit Generation: Automatically generates descriptive commit messages.

  6. Track AI adoption with the Adoption Dashboard

    main

    The Adoption Dashboard allows team leads to monitor how AI is being utilized within the organization. It provides an AI Adoption Score to help quantify usage.

    Key dashboard components include:

    • Overview: Introduction to the AI Adoption Score.
    • Metrics for Team Leads: Specific usage insights and trends.
    • Score Calculation: Documentation on how the adoption score is derived.
    • Improvement Tips: Guidance on how to boost AI adoption within your team.

  7. Automate development workflows with Kilo Code

    main

    Kilo Code allows you to automate repetitive development tasks through several core mechanisms:

    • AI Code Reviews: Automated analysis of Pull Requests (PRs) focusing on Security, Performance, Bug Detection, Style, Tests, and Documentation. You can customize review styles using Strict, Balanced, or Lenient modes.
    • Agent Manager: A control panel to orchestrate multiple AI agents. It supports local and cloud-synced sessions, and features a Parallel Mode that uses Git worktree isolation to run multiple agents simultaneously without workspace conflicts.
    • Model Context Protocol (MCP): A standard to connect Kilo Code to external tools and services via STDIO or SSE transports.
    • Integrations: Native support for GitHub (deployments and code reviews) and GitHub Actions (CI/CD workflows).
    • Extensibility: Support for running local AI models, shell command integration, and auto-launching agents.

  8. Use Kilo's Platform Features beyond the IDE

    main

    Kilo is a multi-platform ecosystem where your Sessions (conversation history, context, and state) sync automatically across all interfaces.

    • Cloud Agents: Run Kilo at app.kilo.ai/cloud to work from any device (tablet/phone) or to offload heavy tasks without using local resources.
    • Parallel Agents: Run multiple agents simultaneously in the IDE or CLI (e.g., one agent writing tests while another handles documentation).
    • App Builder: Build and preview live apps directly in the browser with real-time previews.
    • Kilo Deploy: Perform one-click deployments directly from your workflow.
    • Code Reviews: Automatically analyze PRs for performance, security, style, and test coverage using your chosen AI model.
    • Managed Indexing: Uses cloud-hosted embeddings for semantic search across your repositories for context-aware responses.
    • Autocomplete: Provides inline ghost-text completions with tab-to-complete functionality.

  9. Why Kilo Code uses MCP

    main

    Kilo Code leverages MCP to provide a universal connector between the platform and external services. This approach provides several advantages over direct REST API integrations:

    1. Extensibility: Add unlimited custom tools without waiting for official integrations.
    2. Contextual awareness: Tools can access conversation history and project context.
    3. Simplified integration: Uses one standard protocol rather than numerous unique API patterns.
    4. Runtime flexibility: Discover and use new capabilities on-the-fly.

  10. Understand the ContinueDev library architecture

    main

    The ContinueDev library is a specialized TypeScript service library integrated into Kilocode (located at src/services/continuedev/). It is a streamlined version of the Continue project, focusing exclusively on Autocomplete and NextEdit (edit prediction) functionality. It excludes all GUI, chat, and agent features.

    The system is organized into four distinct layers:

    1. IDE Integration Layer: Adapts the library to specific IDEs (like VSCode or JetBrains) via an IDE interface.
    2. Provider Layer: The main entry points for features, consisting of CompletionProvider (for Autocomplete) and NextEditProvider (for Edit Prediction).
    3. Core Services Layer: Contains shared business logic for LLM communication, context retrieval, and configuration management.
    4. Utility & Infrastructure Layer: Provides low-level services like Tree-sitter (AST), Diff Engines, LRU Caching, and Logging.

  11. Use browser automation and URL ingestion

    main

    Kilo provides two primary non-agent features for interacting with web content:

    1. Browser Automation: Provides browser control to perform deterministic actions and capture screenshots.
    2. URL Ingestion: Converts web content into Markdown format for easier ingestion into AI workflows.

    For detailed feature documentation, refer to the browser-use.md guide in the kilocode-docs application.

  12. External dependencies of the autocomplete module

    main

    The autocomplete module (src/services/autocomplete/) relies on several external and internal layers. When developing or extending this module, be aware of the following dependency categories:

    • VSCode API: Almost all files in the module import * as vscode from "vscode" to interact with the editor environment.
    • Internal Project Layers: The module integrates with src/core/ (e.g., ClineProvider, ProviderSettingsManager, RooIgnoreController), src/api/ (e.g., ApiHandler, OpenRouterHandler), src/i18n (for translations), and src/shared/.
    • Monorepo Packages: It uses @roo-code/types for shared types like ProviderName and AutocompleteServiceSettings, and @roo-code/telemetry for the TelemetryService.
    • Webview UI: It imports constants like PROVIDERS from the webview-ui package to maintain consistency between the autocomplete logic and the settings UI.
    • Third-party & Node.js: The module utilizes approximately 16 third-party npm packages and 8 Node.js built-in modules.