LLM generates text. Agent executes tasks. Partner does research — on its own.
The Problem
As a researcher, I have more ideas than time. Papers to read, experiments to run, code to review, and connections to find between projects. No matter how fast I work, the backlog grows.
What if I had a research companion that worked independently — reading papers, exploring my codebase, building a knowledge base, and proposing new ideas — while I focused on the hard problems?
The Three Layers of AI
We've seen two layers of AI tools:
Introducing Partner 🤝
Partner is an autonomous research entity. It sits on top of existing agent frameworks (like how agents sit on top of LLMs) and conducts research independently.
The core interaction is beautifully simple:
"Hey Partner, what have you been doing?"
And it tells you everything it discovered while you were away.
How It Works
Partner runs in the background, executing a research cycle every 30 minutes (configurable). Each cycle:
- Picks a task from its queue (self-generated or user-injected)
- Executes it via the agent backend (web search, code analysis, etc.)
- Records findings in its knowledge base
- Generates new tasks based on what it learned
- Repeats — forever
Events: The Heart of Partner
An Event is one complete research cycle — like how Agents have Skills, Partner has Events.
Each Event follows a structured flow:
📖 Literature → Search and read papers
🔬 Project Scan → Analyze your codebase
💡 Idea Generate → Propose improvements
🧭 Exploration → Try new directions
📝 Knowledge → Record findings
🌱 Spawn → Create new Events
Events grow on their own — one Event's findings automatically spawn new Events. The research never stops.
Real Results
I ran Partner overnight on my bioinformatics research projects. By morning:
- 29 research cycles completed autonomously
- 34 tasks finished
- 48 knowledge entries accumulated
- 94 tasks queued for future exploration
Key discoveries Partner made on its own:
Multi-Agent Support
Partner works on top of existing agent frameworks — it doesn't reinvent the wheel.
Run partner setup to auto-detect installed agents.
Cross-Platform
Partner runs on Linux, macOS, Windows, and WSL. On WSL, it can automatically access your Windows files through the WSL Bridge.
Getting Started
git clone https://github.com/zty522/partner.git
cd partner
pip install -e .
partner setup
Then open your agent (Hermes, OpenClaw, etc.) and say:
"Hey Partner, what have you been doing?"
What's Next
- WeChat/QQ integration — ask Partner via voice message
- Community Events — share and install Event templates
- Multi-Partner collaboration — multiple Partners working together
- More agent backends — Claude Code, Cursor, and more
Partner: because research shouldn't wait for you.
GitHub:
🤝 Partner
"Hey Partner, what have you been doing?"
An AI research companion that works independently in the background You don't give it commands. You just check in.
The Idea
LLM: You ask → It answers → Done
Agent: You command → It executes → Waits
Partner: It works on its own → You ask "what have you been doing?" → It reports
Partner is proactive. It reads papers, explores your projects, builds a knowledge base, and proposes new ideas — all without you telling it to. When you're ready, you just ask:
"Hey Partner, what have you been doing?"
And it tells you everything it discovered while you were away.
Quick Start
git clone https://github.com/zty522/partner.git
cd partner
pip install -e .
partner setup
The setup wizard detects your installed agents (Hermes, Codex, Claude Code), configures a workspace, and registers Partner as a skill. Then just talk naturally:
You:
