depresearch

Research any open-source repo without leaving your project.

The Problem

You're building an AI chat interface. You see someone on X share their open-source chat app with a slick streaming implementation, or a clever context window optimization, or a feature you want in your app.

Without agents: You go to their repo, read the source code, figure out how the feature works, then manually bring that knowledge back to your project.

With agents: You clone their repo, ask AI how the feature works, copy the response, paste it into the AI chat session that has your project context. Better, but still a lot of friction.

With depresearch: You run one command:

dpr research "how does the streaming implementation work in https://github.com/someone/cool-chat-app"

You get back a walkthrough with file paths, code snippets, and enough context to act on it.

Works as a sub-agent

dpr is a CLI, so your coding agent (Cursor, Claude Code, OpenCode, etc.) can shell out to it directly. The agent gets back what it needs without having to clone and read through a foreign codebase itself.

Your coding agent
    ├── shells out: dpr research "how does X work in <repo-url>"
    │       ├── clones the repo
    │       ├── research agent reads source code
    │       └── returns walkthrough
    └── continues working on your project

Add as an Agent Skill

You can install depresearch as an agent skill so your coding agent knows how to use it automatically:

npx skills add https://github.com/Quintui/depresearch

Once installed, your agent can call dpr on its own whenever it needs to research an external codebase — no manual prompting required.

How It Works

dpr research "your question about a feature in <repo-url>"
    │
    ▼
CLI loads config from ~/.depresearch/
    │
    ▼
Mastra agent starts in-process (no server needed)
    │
    ├── clones repo into ~/.depresearch/workspace/
    ├── reads source code, traces execution paths
    │
    ▼
Detailed walkthrough printed to stdout

The agent runs entirely on your machine using Mastra. It clones repos into ~/.depresearch/workspace/, reads files, runs shell commands — all sandboxed away from your project directory.

Install

npm install -g depresearch

Requires Node.js 20+ and an OpenRouter API key.

Setup

dpr config set api-key <your-openrouter-key>

Run dpr config to see all current settings:

Current configuration:

  api-key   sk-or-v1...a3f2
  model     openrouter/anthropic/claude-haiku-4.5

Usage:
  dpr config set api-key <your-openrouter-key>
  dpr config set model <model-id>
  dpr config get <key>

Usage

# Research a feature in a specific repo
dpr research "how does the streaming response work in https://github.com/mckaywrigley/chatbot-ui"

# Stream response tokens in real-time
dpr research "how do they handle the AI context window in https://github.com/steven-tey/chathn" --stream

# It can also resolve repos by library name
dpr research "how does zod z.infer work internally"

# Change the AI model
dpr config set model openrouter/google/gemini-3-flash-preview

What You Get Back

The research agent traces actual code paths and returns structured walkthroughs:

• Overview — What the feature does
• Key Files & Entry Points — The map of every file and export involved
• Implementation Walkthrough — Step-by-step code trace with generous snippets
• Interfaces & Types — Full type definitions, schemas, API contracts
• Dependencies & Config — External packages, env vars, setup needed
• Patterns & Gotchas — Non-obvious decisions, edge cases, tradeoffs

All responses include file paths with line numbers and code snippets so the output is self-contained — you don't need access to the repo afterward.

Configuration

All config lives in ~/.depresearch/:

~/.depresearch/
├── .env              # OPENROUTER_API_KEY=sk-or-...
├── config.json       # { "model": "openrouter/anthropic/claude-haiku-4.5" }
├── mastra.db         # Agent memory (conversation history)
└── workspace/        # Cloned repos live here

Key	Description	Default
api-key	Your OpenRouter API key	(required)
model	AI model for the research agent (OpenRouter format)	openrouter/anthropic/claude-haiku-4.5

Tech Stack

Component	Technology
CLI framework	citty
AI agent	Mastra (in-process, no server)
LLM provider	OpenRouter
Bundler	tsup
Monorepo	Turborepo + Bun

Development

git clone https://github.com/kristianvtr/depresearch
cd depresearch
bun install
bun run build

# Link globally for local testing
cd apps/cli && npm link

# Watch mode (rebuilds on save)
cd apps/cli && bun run dev

License

MIT