DSRPTV | dsrptv.co

Meta Multi-Agent Coding OS
by DSRPT.AI

Claude Code CLI, but more feature-rich.
Aider + LangGraph + Ollama + Voice + Browser UI
Fully branded Pinokio v7 Terminal Plugin

---

⚡ What is DSRPTV?

DSRPTV is a production-grade meta multi-agent coding system designed to function like Claude Code CLI but with significantly more features:

• 🧠 LangGraph Multi-Agent Core: Supervisor → Coder (Aider) → Tester → Loop with persistent SQLite checkpointing
• 🏛️ Architect Mode: Two-model planning + execution (plan with primary, edit with editor model)
• 🎙️ Voice Input: Local Whisper integration for voice-to-code
• 🌐 Web UI: Aider browser interface (always enabled by default)
• 📊 Auto-Test + Lint: Pytest/npm test + ruff/eslint loops
• 💾 Persistent Memory: Chat history + checkpoint recovery
• 🤖 Ollama Model Picker: Interactive menu to select primary + editor models from your local/remote Ollama instance
• 🚀 Windows Compatible: Direct Windows bash path support via Pinokio kernel

---

📦 Installation

Option 1: Pinokio (Recommended)

1. Open Pinokio
2. Go to Plugins → Install
3. Enter: https://github.com/DSRPT/dsrptv
4. Select your target project directory
5. DSRPTV will auto-install dependencies and launch

Option 2: Manual

git clone https://github.com/DSRPT/dsrptv.git
cd dsrptv
pip install aider-chat langgraph langchain-community langchain-ollama

# Interactive menu launcher
node menu.js

# OR direct LangGraph meta-agent
python graph.py

# OR direct aider launch
bash launch.sh

---

🔧 Usage

Pinokio Plugin Mode

After installation, DSRPTV appears in your Pinokio Plugins list:

1. Click DSRPTV → Select target project
2. Choose environment variables (prompted on first run):
   • OLLAMA_API_BASE: Local (http://localhost:11434) or remote Ollama server
   • DSRPTV_MODEL: Primary model (e.g., ollama/qwen2.5-coder:32b)
   • DSRPTV_EDITOR_MODEL: Editor model (e.g., ollama/qwen2.5-coder:14b)
   • DSRPTV_MODE: architect | agent | code | ask | voice | browser
3. DSRPTV launches directly in your project with Aider Architect mode

Interactive Menu (menu.js)

Run node menu.js for a full interactive setup:

• Ollama Host: Configure local or remote server
• Model Selection: Live-fetch models from Ollama or use recommended list
• Primary + Editor Models: Pick separate models or use same for both
• Agent Mode:
  1. Meta Multi-Agent — LangGraph: Supervisor + Coder + Tester (durable checkpointing)
  2. Native Agent — aider --yes --map autonomous mode
  3. Architect — Two-model plan + execute (default)
  4. Voice-to-Code — Whisper input enabled
  5. Browser Web UI — Aider in browser
  6. Code Mode — Fast single-model editing
  7. Ask Mode — Questions only
• Extras: Voice, persistent memory, auto-test, auto-lint, web URL ingestion

LangGraph Meta-Agent (graph.py)

Run python graph.py for the Supervisor → Coder → Tester loop:

================================================================
  DSRPTV | dsrptv.co — META MULTI-AGENT CODING OS
  by DSRPT.AI — https://dsrpt.ai
  GitHub: https://github.com/DSRPT/dsrptv
================================================================
  LangGraph Meta-Agent Core
  Ollama: http://localhost:11434
  Primary: qwen2.5-coder:32b
  Editor: qwen2.5-coder:14b
  CWD: /your/project
================================================================

[DSRPTV] LangGraph Meta-Agent Ready.
[DSRPTV] Persistent memory: ~/.dsrptv_memory.db
[DSRPTV] Type your coding task below (or 'exit'):

You: add a login route to app.py with JWT authentication

[SUPERVISOR] Analyzing task (iteration 0/5)...
[SUPERVISOR] Routing to CODER

[CODER] Running Aider to implement code changes...
[CODER] Aider succeeded.

[SUPERVISOR] Routing to TESTER

[TESTER] Running tests and lint...
[TESTER] pytest passed.
[TESTER] ruff passed.

[SUPERVISOR] Code applied and tests passed. Task complete.

[DSRPTV] Task complete. State persisted to ~/.dsrptv_memory.db

Direct Aider Launch (launch.sh)

Run bash launch.sh for immediate Aider Architect mode with DSRPTV branding:

================================================================
  DSRPTV | dsrptv.co — META MULTI-AGENT CODING OS
  by DSRPT.AI — https://dsrpt.ai
  GitHub: https://github.com/DSRPT/dsrptv
================================================================
  Model: ollama/qwen2.5-coder:32b
  Editor: ollama/qwen2.5-coder:14b
  Mode: architect
  Ollama: http://localhost:11434
================================================================

aider --model ollama/qwen2.5-coder:32b --editor-model ollama/qwen2.5-coder:14b --architect --auto-accept-architect --yes --stream --map --message-history .dsrptv-memory.json

---

🎯 Features Comparison

Feature	Claude Code CLI	DSRPTV
Terminal Interface	✔️	✔️
Architect Mode	✔️ (single model)	✔️ (dual model: primary + editor)
Autonomous Agent	✔️	✔️
LangGraph Multi-Agent	❌	✔️ (Supervisor + Coder + Tester)
Persistent Checkpointing	❌	✔️ (SQLite)
Ollama Support	❌	✔️ (local + remote)
Model Picker	❌	✔️ (interactive)
Voice Input	❌	✔️ (local Whisper)
Auto-Test + Lint Loops	❌	✔️ (pytest/npm, ruff/eslint)
Web URL Ingestion	✔️	✔️
Pinokio Plugin	❌	✔️ (fully integrated)
Windows Support	Limited	✔️ (native bash path)
Open Source	❌	✔️

---

📁 Repository Structure

DSRPT/dsrptv/
├── README.md          # This file
├── pinokio.js         # Pinokio v7 Terminal Plugin config
├── launch.sh          # Direct Aider launcher with DSRPTV branding
├── menu.js            # Interactive Node.js launcher (Ollama picker, mode selection, extras)
├── graph.py           # LangGraph meta-agent core (Supervisor + Coder + Tester)
└── icon.png           # DSRPTV icon (optional)

---

🧩 Architecture

1. Pinokio v7 Plugin (pinokio.js)

• Env Prompts: First-run config for Ollama host, models, mode
• Windows Bash: Direct kernel path to miniconda/Library/bin/bash.exe
• Auto-Install: pip install aider-chat langgraph ... on first run
• Direct Launch: Executes aider in target project ({{args.cwd}})

2. Interactive Menu (menu.js)

• Ollama API: Fetches live model list from /api/tags
• Model Selection: Numbered menu with defaults
• 7 Agent Modes: Meta multi-agent, native agent, architect, voice, browser, code, ask
• Extras: Voice, memory, auto-test, auto-lint, web URLs
• Command Builder: Constructs full aider command with all flags

3. LangGraph Meta-Agent (graph.py)

StateGraph Workflow:

      START
        ↓
   [SUPERVISOR]
        │
    ┌───┼───┐
    │       │
 [CODER] [TESTER]
    │       │
    └───┬───┘
        │
   [SUPERVISOR]
        ↓
    FINISH/LOOP

• Supervisor Node: Routes to coder/tester based on code_applied and tests_passed state
• Coder Node: Runs aider --model ollama/{EDITOR_MODEL} --yes --auto-commits --message "{task}"
• Tester Node: Runs pytest + ruff, updates state
• Checkpointing: SqliteSaver persists state to ~/.dsrptv_memory.db
• Max Iterations: 5 (configurable)

4. Direct Launcher (launch.sh)

• ASCII Banner: DSRPTV branding
• Auto-Install: Checks for aider, installs if missing
• Direct Exec: aider --architect --auto-accept-architect ...

---

👥 Credits

Built by DSRPT.AI
Website: dsrptv.co
GitHub: DSRPT/dsrptv

Powered By

• Aider — AI pair programming in terminal
• LangGraph — Multi-agent orchestration
• Ollama — Local LLM inference
• Pinokio — One-click AI app launcher

---

📜 License

MIT License - see LICENSE for details.

---

🔗 Links

• Website: dsrptv.co
• DSRPT.AI: dsrpt.ai
• GitHub: DSRPT/dsrptv
• Issues: Report a bug