The Agent. The System. The Solution.

Knox is an AI-powered VS Code extension that combines an autonomous agent, context-aware code completion, automatic checkpoints, and a full suite of built-in tools into a single development environment. It supports Anthropic Claude, OpenAI GPT, and Knox's own models out of the box.

Core Features

AI Chat Interface

A sidebar chat powered by your choice of LLM. Conversations are streamed in real time with full tool execution displayed inline.

• Multi-model support — Anthropic Claude (4.5/4.6 Haiku, Sonnet, Opus), OpenAI GPT (GPT-5.4, GPT-5.4 Pro), and Knox's memory system On-Demand model knox/knox-ms
• Role-based model routing — Assign different models to different tasks: chat, autocomplete, edit, apply, summarize, viewRead, realTimeSearch
• Session management — Multiple conversation tabs, session history, and session restore
• Rich context — Attach files, folders, selections, images/screenshots, terminal output, git diffs, and 26 context providers

Agent Mode

An autonomous agent that can plan, execute, and verify multi-step development tasks.

• Task decomposition — AI breaks complex requests into a structured todo list with priorities, dependencies, and complexity estimates
• Built-in tools — The agent has access to 14+ tools: file read/create, terminal commands, code search, web search, repo map, git diff, LSP operations (go-to-definition, find references, hover, symbols), deep code analysis, and multi-file semantic search
• Shadow workspace — Test changes in an isolated environment before applying
• Auto-verification — Automatically verifies changes after applying (runs diagnostics, checks for errors)
• Error pattern detection — Monitors terminal output for 20+ error patterns and suggests fixes
• Undo/redo — Cmd/Ctrl + Shift + Alt + Z to undo, Cmd/Ctrl + Shift + Alt + Y to redo agent operations
• Diagnostic integration — Check and fix code diagnostics with AI assistance

Code Completion

Tab-triggered inline autocomplete that understands your codebase context.

• Context-aware suggestions — Uses recently edited files, visited ranges, and project structure
• Battery-aware — Can pause on low battery to conserve power
• Toggle — Cmd/Ctrl + K, Cmd/Ctrl + A to enable/disable
• Status bar indicator — Shows Enabled, Disabled, or Paused state

Inline Code Editing

Natural language code edits with visual diff review.

• Cmd/Ctrl + I — Describe a change in English; Knox generates and streams a diff
• Vertical inline diffs — See proposed changes directly in the editor with green/red highlighting
• Accept/Reject per block — Alt + Cmd/Ctrl + Y to accept, Alt + Cmd/Ctrl + N to reject individual blocks
• Accept/Reject all — Shift + Cmd/Ctrl + Enter to accept all, Shift + Cmd/Ctrl + Backspace to reject all
• Multi-file batch diffs — Review and apply changes across multiple files in a dedicated Batch Diff panel
• Smart apply strategies — Unified diff detection, LLM-assisted lazy apply, or AST-based fallback

Checkpoint System

Automatic version control for AI-generated changes. Never lose work.

• Auto-checkpoints — Created before every AI conversation response
• Manual checkpoints — Create on demand via the command palette
• One-click restore — Revert to any previous checkpoint instantly
• Configurable retention — Max checkpoints (default: 1000), retention period (default: 7 days), max storage (default: 1 GB)
• Compression — Enabled by default to minimize storage
• File filtering — Tracks 22+ file extensions (js, ts, py, java, go, rs, cpp, etc.) and respects .knoxignore
• Conflict resolution — Handles checkpoint conflicts gracefully
• Visual timeline — Browse checkpoints from the sidebar title bar or command palette
• Enterprise monitoring — Built-in metrics and performance monitoring for checkpoint operations

Slash Commands

Built-in slash commands for common workflows:

Custom slash commands can be defined in config.yaml.

Context Providers

Knox provides 26 context providers that can be attached to any conversation using @:

Context Menu Actions

Right-click selected code to access Knox actions:

• Add as Context — Send selection to Knox chat
• Write Comments — Generate inline comments
• Write Docstring — Generate documentation
• Fix Code — AI-powered code fix
• Optimize Code — Suggest optimizations
• Fix Grammar / Spelling — For Markdown files

Explorer context menu:

• Select Files as Context — Add files to chat context

Terminal context menu:

• Debug Terminal — AI-assisted terminal debugging

Rules System

Define project-specific coding standards that are injected into every AI interaction.

File locations (merged by priority, highest last):

• ~/.knoxrules — Global rules
• {workspace}/.knoxrules — Workspace-level rules
• {workspace}/.knox/rules/*.md — Per-topic rule files
• {subfolder}/.knoxrules — Folder-level overrides

Format:

---
applyTo: ["**/*.ts", "**/*.tsx"]
priority: 10
---
- Use `interface` for object shapes, `type` for unions
- Prefer `const` over `let`
- All exports must have JSDoc

Supports applyTo glob patterns, priority levels, and template variables ({os}, {arch}, {home}).

Skills System

Reusable domain-specific instruction sets that extend Knox's capabilities.

Discovery locations:

• .knox/skills/, .claude/skills/, .agents/skills/ (project-level)
• ~/.knox/skills/, ~/.claude/skills/, etc. (global)

Each skill is a folder containing a SKILL.md:

---
name: react-components
description: React component development patterns
---
## Instructions
- Use functional components with hooks
- Extract shared logic into custom hooks
@src/components/Button.tsx

List loaded skills with /skills in chat.

Prompt Files

Custom reusable prompts stored as .prompt files in .prompts/ or .knox/prompts/.

---
name: api-endpoint
description: Generate a new API endpoint
---
Create a RESTful endpoint following our patterns.
@src/routes/example.ts
@currentFile
<system>You are an expert backend developer.</system>

Supports file references (@file.ts), URL references (@https://...), context provider references (@currentFile, @repo-map), system message blocks, and recursive nesting.

Configuration

Knox can be configured via config.yaml (in .knox/ or workspace root) or the interface settings.

VS Code Settings

Keyboard Shortcuts

Language & File Support

Knox works across all major languages and file types:

• JavaScript / TypeScript (.js, .jsx, .ts, .tsx)
• Python (.py)
• Java (.java) / Kotlin (.kt)
• C / C++ (.c, .cpp)
• C# (.cs)
• Go (.go)
• Rust (.rs)
• PHP (.php)
• Ruby (.rb)
• Swift (.swift)
• HTML / CSS / SCSS (.html, .css, .scss)
• JSON / YAML (.json, .yaml, .yml)
• Markdown (.md)
• SQL (.sql)
• Custom .prompt files — Syntax-highlighted prompt language

JSON schema validation is provided for .knox/config.json and .knoxrc.json.

Internationalization

Knox supports multiple languages:

• English (default)
• Chinese (中文)

Language can be changed from the Settings page in the sidebar.

Project Memory

A SQLite-backed persistent memory store for project-specific knowledge.

• Store conventions, architectural decisions, and patterns
• Memory is searchable and automatically surfaced as context
• Manage via the Memory panel in the sidebar
• Create, search, and delete memory entries

MCP Server Support

Knox integrates with the Model Context Protocol for extensible tool and context access.

• Configure MCP servers in .knox/mcp.json in your workspace directory
• Click "Add MCP Servers" in the MCP panel to open the config file directly
• MCP tools appear as agent tools automatically
• MCP context providers are available via @MCP
• Dynamic slash commands from MCP servers
• Config auto-reloads on save — servers connect seamlessly

Example .knox/mcp.json:

{
  "mcpServers": [
    {
      "name": "filesystem",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem"],
      "env": {}
    }
  ]
}

Requirements

• VS Code ≥ 1.109.0
• Node.js ≥ 22.22.0
• An API key for at least one supported LLM provider (Anthropic, OpenAI, or Knox)

Documentation & Support

• Docs: https://docs.knox.chat/
• GitHub: https://github.com/knoxchat/knoxchat
• Email: support@knox.chat
• Homepage: https://knox.chat