BossAgent

The AI Workforce OS for founders, operators, and ambitious builders.

Configure a team of 24/7 AI employees to research, plan, write, analyze, follow up, and produce execution-ready reports.

English · 简体中文 · 繁體中文 · 日本語 · 한국어 · Español · Português · Français

---

Project Introduction

BossAgent is a local-first, extensible AI Agent workspace that helps founders, operators, creators, and teams configure a coordinated team of AI employees.

It is not another chatbot. BossAgent turns business goals into multi-agent workflows: opportunity research, project judgment, scenario simulation, data analytics, content generation, sales follow-up, document analysis, and final executive reporting.

Why BossAgent

AI should not just chat. It should work.

Founders do not need another empty conversation box. They need an AI team that can split a goal into specialized work, keep context, and return a useful operating report.

BossAgent is designed as a practical MVP for local demos, commercial experiments, and future integrations with tools such as CrewAI, Dify, n8n, WeCom, Feishu, and browser automation.

Core Agents

Agent	Status	Responsibility
Chief of Staff Agent	Available	Decomposes tasks, coordinates specialist agents, and summarizes the final report.
Opportunity Radar Agent	Available	Finds trends, industry windows, opportunity signals, required resources, and risks.
Strategy Advisor Agent	Available	Judges whether a project is worth entering and designs low-cost validation paths.
Content Clone Agent	Available	Drafts social posts, Xiaohongshu content, short-video scripts, and livestream talk tracks.
Sales Follow-up Agent	Available	Creates customer segmentation, follow-up scripts, quote communication, and deal motion.
Document Analyst Agent	Available	Reads PDF, Word, Excel, CSV, and TXT files, then extracts summaries and action lists.
Scenario Simulation Agent	Available	Local MiroFish-inspired sandbox for multi-role reaction simulation, risk discovery, and decision stress testing.
Data Analytics Agent	Available	Local Data Analytics-inspired workflow for data quality, metric summaries, chart previews, and business recommendations.

Features

Feature	Status
Multi-agent workflow orchestration	Available
Streamlit local UI	Available
OpenAI-compatible model support	Available
Local GGUF model folder mode	Experimental
File upload and document analysis	Available
CSV / Excel data analytics page	Available
MiroFish-style scenario simulation task	Available
Dedicated MiroFish x Data Analytics drill lab	Available
SQLite task history	Available
Markdown / DOCX report export	Available
CrewAI-native orchestration	Planned
Browser automation and external actions	Planned, human confirmation required
Team workspace and permissions	Planned

Screenshots / Preview

Screenshots coming soon.

• Dashboard Preview
• Multi-agent Workflow Preview
• Report Output Preview

Quick Start

git clone https://github.com/smclw/BossAgent.git
cd BossAgent
python -m venv .venv

Windows PowerShell:

.\.venv\Scripts\python.exe -m pip install -r requirements.txt
Copy-Item .env.example .env
.\.venv\Scripts\python.exe -m streamlit run app.py

If you want to activate the environment in PowerShell:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass
.\.venv\Scripts\Activate.ps1

macOS / Linux:

source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
streamlit run app.py

Open the local URL printed by Streamlit, usually http://localhost:8501.

Run on Google Colab

BossAgent can also be launched from Google Colab for quick demos. Colab is best for testing the UI and mock workflows; avoid uploading private customer documents or hard-coding API keys in shared notebooks.

1. Clone and install

!git clone https://github.com/smclw/BossAgent.git
%cd BossAgent
!pip -q install -r requirements.txt
!cp .env.example .env

2. Start in mock mode

!printf "\\nLLM_PROVIDER=mock\\nUSE_MOCK_LLM=true\\n" >> .env
!streamlit run app.py --server.port 8501 --server.address 0.0.0.0 > streamlit.log 2>&1 &

3. Expose the Streamlit page

Colab cannot open localhost:8501 directly from your browser. Use a temporary tunnel:

!wget -q https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64 -O cloudflared
!chmod +x cloudflared
!./cloudflared tunnel --url http://localhost:8501

Open the generated https://...trycloudflare.com URL. Keep the Colab cell running while using the app.

Optional: use an API key safely

For real model calls, store your key in Colab Secrets instead of writing it into the notebook:

from google.colab import userdata
api_key = userdata.get("OPENAI_API_KEY")

from pathlib import Path
env_path = Path(".env")
text = env_path.read_text()
text += f"\\nLLM_PROVIDER=openai-compatible\\nUSE_MOCK_LLM=false\\nOPENAI_API_KEY={api_key}\\n"
env_path.write_text(text)

Optional: run with a downloaded local GGUF model

You can also run BossAgent in Colab without an API key by downloading a .gguf model into the models/ folder. The default example below uses Qwen3-14B Q4_K_M, which is much more useful than toy-size models but needs more memory. For Colab, use a GPU or high-RAM runtime when possible.

!pip -q install -r requirements-local.txt
!mkdir -p models

# Qwen3 14B GGUF model for more realistic local demos.
# You can replace it with another direct GGUF URL.
GGUF_URL = "https://huggingface.co/Qwen/Qwen3-14B-GGUF/resolve/main/Qwen3-14B-Q4_K_M.gguf"
!wget -O models/local-model.gguf "$GGUF_URL"

from pathlib import Path
env_path = Path(".env")
text = env_path.read_text()
text += "\\nLLM_PROVIDER=local-folder\\nUSE_MOCK_LLM=false\\nLOCAL_MODEL_DIR=models\\nLOCAL_GPU_LAYERS=0\\n"
env_path.write_text(text)

Then start Streamlit with the same command and open it through the temporary tunnel. In the app sidebar, choose Local GGUF model folder if you want to switch modes after startup.

Note: Qwen3-14B Q4_K_M is a multi-GB model. If Colab runs out of memory, switch to a high-RAM/GPU runtime or replace the URL with a smaller GGUF model.

Configuration

BossAgent reads model and runtime settings from .env.

LLM_PROVIDER=mock
OPENAI_API_KEY=
MODEL_NAME=gpt-4.1-mini
API_BASE_URL=https://api.openai.com/v1
USE_MOCK_LLM=true
TEMPERATURE=0.4
LOCAL_MODEL_DIR=models

Supported model paths:

• OpenAI-compatible APIs: OpenAI, DeepSeek, Qwen DashScope compatible mode, Ollama, LM Studio, and similar endpoints.
• Local folder mode: experimental direct loading of .gguf models via llama-cpp-python.

For direct local GGUF loading:

.\.venv\Scripts\python.exe -m pip install -r requirements-local.txt

Then put a .gguf model into models/ or set LOCAL_MODEL_DIR to a full .gguf file path.

At app startup, the sidebar includes a Model startup mode selector. You can switch between mock mode, OpenAI-compatible API, and a local GGUF model folder without editing files manually.

For local GGUF mode, BossAgent defaults to LOCAL_CONTEXT_WINDOW=16384 and automatically compresses long prompts before calling llama-cpp-python. If you still see context-window errors, increase LOCAL_CONTEXT_WINDOW in .env, lower LOCAL_MAX_TOKENS, or use a model/runtime with more memory.

MiroFish x Data Analytics Drill Lab

The Drill Lab is a standalone local module that combines two workflows:

• Data Analytics layer: reads CSV / Excel files locally, builds a data profile, checks missing values, summarizes numeric and categorical fields, and previews simple charts.
• MiroFish-inspired simulation layer: uses the data profile as evidence, then simulates multi-role reactions from customers, sales, delivery, competitors, channels, investors, and public opinion.

It is designed for business rehearsals such as launches, pricing, sales motions, channel expansion, and decision stress tests. It does not execute external actions and should be treated as structured scenario rehearsal, not guaranteed prediction.

Roadmap

• v0.1 Local MVP: multi-agent workflow, task input, file parsing, history, and report export. Available.
• v0.2 MiroFish-style scenario simulation and local Data Analytics workflow. Available.
• v0.3 Knowledge base and long-term memory. Planned.
• v0.4 Browser automation, email, spreadsheets, Feishu, WeCom, Dify, and n8n connectors. Planned.
• v0.5 Team workspace, multi-user mode, and permissions. Planned.
• v1.0 Deployable AI Workforce OS. Planned.

Contributing

Contributions are welcome:

• Build new agents.
• Improve workflows.
• Add model providers.
• Improve the UI.
• Add connectors.
• Translate documentation.
• Report bugs and propose product ideas.

Please read CONTRIBUTING.md before opening a pull request.

Security

• Never commit API keys.
• Never upload private customer data.
• Do not execute uploaded files.
• External actions such as sending, deleting, paying, ordering, or modifying third-party systems must require human confirmation.

See SECURITY.md for details.

License

BossAgent is released under the MIT License.