🔬 Advanced Features

Power user features for getting the most out of Code Puppy.

🛡️ Privacy Policy — Verified Zero Telemetry

✅ What We DON'T Collect

Data Type	Status	Verification
Telemetry	ZERO	MITM integration tests
Prompt Logging	ZERO	No server-side storage
Behavioral Profiling	ZERO	No analytics code
Third-Party Data Sharing	ZERO	Network traffic analysis
Usage Analytics	ZERO	No tracking endpoints

🔬 How We Verify: MITM Integration Tests

Code Puppy's test suite includes Man-in-the-Middle (MITM) integration tests that:

1. Intercept all network traffic between the agent and LLM providers
2. Analyze every HTTP request to verify only legitimate API calls are made
3. Detect any telemetry from open-source dependencies that might phone home
4. Fail the build if any unexpected network activity is detected

This isn't a privacy policy document—it's enforced by automated testing on every release.

🏠 Privacy by Architecture

🤷 Why Other Tools Can't Promise This

The AI coding tool landscape is crowded, but nearly every tool has external pressures that Code Puppy simply doesn't:

Tool	Backing	Source	Concern
Cursor	VC-funded	Closed	Investor pressure to monetize
Windsurf	VC-funded	Closed	Investor pressure to monetize
Opencode	VC-funded	Open	Investor pressure to monetize
Crush	VC-funded	Open	Investor pressure to monetize
Copilot	Corporate (Microsoft)	Closed	Serves corporate interests
Claude Code	Corporate (Anthropic)	Open	Serves corporate interests
Roo Code	VC-funded (fork)	Open	Investor pressure to monetize
Kilo Code	VC-funded (fork)	Open	Investor pressure to monetize

The Pattern:

• 💰 VC-funded tools have investors expecting returns—your data is the product
• 🏢 Corporate-backed tools serve corporate interests first, users second
• 🔀 Fork ecosystems have unclear provenance—who's actually maintaining the code?
• 🔒 Closed source tools can't be audited—you're trusting their word
• 📄 Privacy policies can change whenever business needs change

Code Puppy is Different:

• ✅ No investors – No pressure to monetize your data
• ✅ No corporate overlords – We answer to users, not shareholders
• ✅ Truly open source – Audit every line of code yourself
• ✅ Verified by tests – MITM integration tests prove our claims
• ✅ Community-driven – Built for developers, by developers

Agent Rules (AGENT.md)

Create an AGENT.md file in your project root to give Code Puppy project-specific context and rules:

# AGENT.md

## Project: My Awesome App

### Overview
This is a FastAPI backend with a React frontend.

### Coding Standards
- Use TypeScript for all frontend code
- Python backend uses type hints everywhere
- Follow PEP 8 for Python, Prettier for JS/TS

### Architecture
- Backend: FastAPI with SQLAlchemy
- Frontend: React with TypeScript
- Database: PostgreSQL

### Testing
- Backend: pytest with 80% coverage
- Frontend: Jest + React Testing Library

### Important Files
- src/api/main.py - FastAPI entry point
- frontend/src/App.tsx - React entry point

Custom Commands

Create project-specific slash commands that Code Puppy will automatically discover. Commands are looked for in three locations:

• .claude/commands/ – Claude Code compatible
• .github/prompts/ – GitHub Copilot compatible
• .agents/commands/ – Code Puppy native

Create a markdown file in any of these directories:

# .agents/commands/deploy.md
## Deploy to Production

1. Run all tests
2. Build the production bundle
3. Deploy to AWS
4. Verify health checks

Then use with /deploy in your prompts. The command name comes from the filename!

File Attachments

Reference files in your prompts using @:

> Review @src/main.py and suggest improvements

> Compare @old/config.yaml with @new/config.yaml

> Refactor @utils.py to use async/await

Use Tab for path auto-completion!

Round-Robin Models

Distribute requests across multiple API keys or providers to avoid rate limits. Configure in ~/.code_puppy/extra_models.json:

{
  "qwen1": {
    "type": "cerebras",
    "name": "qwen-3-coder-480b",
    "custom_endpoint": {
      "url": "https://api.cerebras.ai/v1",
      "api_key": "$CEREBRAS_API_KEY1"
    },
    "context_length": 131072
  },
  "qwen2": {
    "type": "cerebras",
    "name": "qwen-3-coder-480b",
    "custom_endpoint": {
      "url": "https://api.cerebras.ai/v1",
      "api_key": "$CEREBRAS_API_KEY2"
    },
    "context_length": 131072
  },
  "qwen3": {
    "type": "cerebras",
    "name": "qwen-3-coder-480b",
    "custom_endpoint": {
      "url": "https://api.cerebras.ai/v1",
      "api_key": "$CEREBRAS_API_KEY3"
    },
    "context_length": 131072
  },
  "cerebras_round_robin": {
    "type": "round_robin",
    "models": ["qwen1", "qwen2", "qwen3"],
    "rotate_every": 5
  }
}

The rotate_every setting controls how many requests are sent to each model before rotating to the next. In this example, every 5 requests it switches to the next API key.

Use cases:

• Avoiding rate limits by spreading requests across multiple API keys
• Load balancing across providers
• Cost optimization by mixing expensive/cheap models

Multi-Agent Workflows

Agents can invoke other agents using the invoke_agent tool. This enables powerful delegation patterns where a coordinator (like the planning agent) can dispatch work to specialists.

Parallel Sub-Agents

If your model supports parallel tool calls, it can invoke multiple sub-agents simultaneously! Models with parallel tool calling support:

• ✅ Claude (all claude-4-5 models)
• ✅ OpenAI (gpt-5.1, gpt-5.2)
• ✅ Google (Gemini-3)

Example: Planning Agent Delegation

The planning agent can break down a complex task and delegate to specialists in parallel:

# Planning agent receives: "Add authentication to my app"
# It might invoke in parallel:

invoke_agent("security-auditor", "Audit current auth setup")
invoke_agent("python-programmer", "Implement JWT middleware")
invoke_agent("qa-expert", "Design auth test strategy")

Each sub-agent works independently, and results are collected by the coordinator. This dramatically speeds up complex multi-step tasks!

DBOS Durable Execution

For long-running tasks, Code Puppy can use DBOS for durable execution:

• Automatic checkpointing
• Resume after interruption
• Transaction guarantees

Keyboard Shortcuts

Shortcut	Action
Ctrl+C	Cancel current operation
Ctrl+D	Exit Code Puppy
Tab	Auto-complete commands/paths
@	Trigger file path completion
Up/Down	Navigate command history
Ctrl+R	Reverse search history

Environment Detection

Code Puppy automatically detects:

• Git repository context
• Project type (Python, Node, etc.)
• Available tools (npm, pip, cargo, etc.)
• CI/CD environment

Pro Tips