GitHub - opencolin/openengineer: 12 specialized AI agents for OpenCode — one per AI lab (Qwen, DeepSeek, NVIDIA, Google, Meta, Moonshot, Z.ai, NousResearch, OpenAI, MiniMax, PrimeIntellect). Cross-session memory, cost tracking, ZSH shell plugin. All open-source via Nebius Token Factory. · GitHub
Skip to content

opencolin/openengineer

 
 

Repository files navigation

Open Engineer

A batteries-included multi-model AI agent team for OpenCode. 12 specialized agents, each powered by the best open-source model from a different AI lab.

Forked from oh-my-openagent by code-yeongyu.

Agents

One agent per model maker. No duplicates.

Agent Primary Model Maker Role
architect Qwen3 Coder 480B Qwen Main orchestrator
engineer GPT-OSS 120B OpenAI Autonomous deep worker
strategist DeepSeek R1 DeepSeek Deep reasoning
librarian MiniMax M2.5 MiniMax OSS code & docs search
researcher Nemotron Ultra 253B NVIDIA Broad knowledge synthesis
analyst Gemma 3 27B Google Fast codebase exploration
designer Qwen2.5 VL 72B Qwen* Visual/UI tasks
product-manager Kimi K2.5 Moonshot Strategic planning
consultant GLM-5 Z.ai Architecture advisor
qa-engineer Hermes 4 405B NousResearch Plan reviewer
technical-lead Llama 3.3 70B Meta Plan executor
junior-architect INTELLECT-3 PrimeIntellect Lighter coding tasks

*Qwen2.5 VL 72B is the only vision-capable model available — retained for designer.

Categories

Category Primary Model Use
visual-engineering Qwen2.5 VL 72B Frontend, UI, design
ultrabrain DeepSeek R1 Hard logic-heavy tasks
deep Kimi K2.5 Autonomous problem-solving
artistry Nemotron Ultra 253B Creative approaches
quick Gemma 3 27B Trivial tasks
unspecified-low Llama 3.3 70B Moderate unclassified tasks
unspecified-high Qwen 3.5 397B High-effort unclassified tasks
writing INTELLECT-3 Documentation and prose

Features

  • 12 Specialized Agents — one per AI lab
  • Cross-Session Memory — persistent facts via mem0 (local, no cloud)
  • Cost Tracking — real-time per-session and lifetime cost in USD
  • ZSH Shell Plugin — invoke agents from terminal with : prefix
  • All Open Source — no proprietary model lock-in

Installation

npm install openengineer

Add to opencode.json:

{
  "plugin": ["openengineer"],
  "model": "nebius/deepseek-ai/DeepSeek-V3.2",
  "provider": {
    "nebius": {
      "name": "Nebius Token Factory",
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.tokenfactory.us-central1.nebius.com/v1"
      }
    }
  }
}
export NEBIUS_API_KEY="your-key"

Shell Plugin

./shell-plugin/install.sh

: fix the login bug
:strategist should we use Redis or Memcached?
:researcher what are the best approaches for distributed caching?
:plan migrate the database

License

See LICENSE.md.

About

12 specialized AI agents for OpenCode — one per AI lab (Qwen, DeepSeek, NVIDIA, Google, Meta, Moonshot, Z.ai, NousResearch, OpenAI, MiniMax, PrimeIntellect). Cross-session memory, cost tracking, ZSH shell plugin. All open-source via Nebius Token Factory.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

Contributors

No contributors

Languages

  • TypeScript 95.0%
  • HTML 4.1%
  • MDX 0.5%
  • Python 0.2%
  • Shell 0.1%
  • JavaScript 0.1%