Cappy

The open-source AI agent that turns your IDE into an autonomous development environment.
Multi-agent swarms · Semantic code search · Persistent memory · Human-in-the-loop safety

The Problem

You're deep in a coding session. New work arrives from multiple channels. You context-switch between issue tracker, terminal, docs, and the editor. Flow is lost.

What if an agent team inside your IDE could plan, execute, and validate tasks — with full project context and your approval?

The Solution: Meet Cappy

Cappy is not just another coding assistant. It's a full autonomous agent runtime living inside VS Code — with multi-agent swarms, semantic code search, persistent project memory, and human-in-the-loop safety built in from day one.

Pick any model on OpenRouter. Extend with MCP servers. Stay in control.

Key Features

Multi-Agent Swarms

One agent is good. A coordinated team is unstoppable.

Cappy's Coordinator Mode decomposes complex tasks into independent sub-tasks and delegates them to parallel worker agents — automatically.

Agents communicate through a mailbox-based swarm protocol with broadcast and point-to-point messaging.

Semantic Code Search (RAG)

Search by meaning, not just text.

Cappy indexes your codebase with AI embeddings and understands your architecture:

• Git-aware incremental indexing — only re-indexes changed files
• AST-powered dependency graphs — understands imports, inheritance, and module structure
• Natural language queries — "where is user authentication handled?"
• Configurable embedding model — text-embedding-3-small by default

25+ Built-in Tools

Everything you need without leaving the chat.

Persistent Project Memory

Cappy gets smarter the more you use it.

The memory system stores structured knowledge in .cappy/memory/ — project overview, architecture decisions, conventions, known pitfalls, and active workstreams. Memory is automatically injected into every session.

Workspace Skills

Teach Cappy your team's playbooks.

Create markdown skill files in .cappy/skills/ to share project-specific workflows. Skills are automatically discovered and offered to the agent as domain knowledge — perfect for onboarding, deployment runbooks, or custom code generation patterns.

Human-in-the-Loop Safety

Full autonomy when you want it. Full control when you need it.

Every destructive operation requires your explicit approval:

• confirm_each (default) — pause before every file write, shell command, or MCP call
• allow_all — trust mode for experienced users

Plan Mode adds an extra guard: the agent reasons freely, but all mutations are blocked until you exit planning.

Smart Context Management

Long conversations, handled gracefully.

• 128K token context window with automatic trimming
• LLM-based compression — dropped messages are summarized and re-injected
• Tool argument recovery — malformed JSON is auto-repaired

Three Chat Modes

MCP Protocol Support

Extend Cappy with any tool server.

Connect multiple Model Context Protocol servers simultaneously. SSE and Stdio transports supported. Automatic destructive-tool detection. Sandboxed by chat mode.

OpenClaude CLI Integration

Launch the OpenClaude CLI directly from VS Code with workspace-aware configuration. Switch seamlessly between Cappy's visual dashboard and the terminal CLI.

Live Dashboard

Everything at a glance.

A sleek sidebar dashboard built with React + Mantine:

• Streaming responses with live progress and elapsed time
• HITL confirmation cards with approve/reject in one click
• Session history with search, pin, and archive
• Tool-call timeline showing every action taken
• Inline file diffs for every code change
• Config & MCP panels for quick settings access

Getting Started

Install

Search for "Cappy" in the VS Code Marketplace, or:

code --install-extension eduardocecon.cappy

Configure

1. Open the Cappy sidebar (Activity Bar icon)
2. Go to the Config tab
3. Enter your OpenRouter API key
4. Choose a model and start chatting

Example Workflows

Feature implementation with agent swarm

You: "Implement JWT authentication with refresh tokens.
      Use a coordinator to split the work."
Cappy: [spawns workers for token service, middleware, and tests — 
        runs them in parallel, merges results]

Semantic code search

You: "Find all the places where we handle payment errors"
Cappy: [uses RAG embeddings to locate semantically relevant code]

Architecture planning

You: "Plan the migration from REST to GraphQL for the user module"
Cappy: [enters Plan Mode, analyzes dependencies, generates step-by-step plan]

Configuration Reference

Development

git clone https://github.com/cecon/cappy.git
cd cappy
pnpm install
pnpm dev       # Start webview dev server
# Press F5 in VS Code to launch Extension Development Host

License

MIT — see LICENSE