Auto-Analyst — Your Open-Source AI Data Scientist

By Firebird Technologies

Auto-Analyst is a fully open-sourced, modular AI system designed to automate data science workflows — from data cleaning and statistical analysis to machine learning and visualization.

You can try it live at: https://www.autoanalyst.ai/chat

---

🚀 Highlights

• ✅ Open Source: Licensed under a highly MIT permissive license.
• 🔄 LLM Agnostic: Compatible with any LLM API – OpenAI, Anthropic, Deepseek (groq), etc.
• 💸 Bring Your Own API Key: No vendor lock-in; use your own keys, pay only what you use.
• 🖥️ User-Centric UI: Built with data scientists in mind.
• 🛡️ Reliable Outputs: Guardrails for robust and interpretable responses.
• ⚙️ Modular Agent Architecture: Add or customize agents using DSPy.

---

Live App

Start analyzing here: 👉 https://www.autoanalyst.ai/chat

---

How It Works

🪜 Step-by-Step Walkthrough

1️⃣ Upload Your Dataset

• Click the 📎 icon near the chat input.
• Upload .csv or .xlsx files. More connectors (APIs, SQL, etc.) available upon request.

2️⃣ Describe Your Dataset

• Enter a short text description of what your dataset is about.
• Auto-Analyst will generate a cleaned, structured metadata summary optimized for LLM workflows.
• ✍️ Tip: Rename generic columns like var_1 to price, category, etc., for better analysis.

3️⃣ Ask a Question

Use either:

• @agent_name to specify which agent to use (e.g. @preprocessing_agent)
• Or no agent tag to let the planner route your query automatically.

---

Built-in Agents

Agent	Description
@preprocessing_agent	Cleans data using pandas and numpy. Fixes types, handles nulls, computes aggregates.
@statistical_analytics_agent	Performs regression, correlation, ANOVA, and other statistical tests with statsmodels.
@sk_learn_agent	Trains machine learning models like Random Forest, KMeans, Logistic Regression using scikit-learn.
@data_viz_agent	Generates visualizations using plotly. Includes a retriever to pick optimal chart formats.

🌟 Modular and extensible! You can add custom agents for:

• Marketing
• Quantitative Finance
• Web APIs (Slack, Notion, etc.)

---

💬 Planner Mode

Want to delegate the query routing?

Just type your question without specifying an agent. The planner will:

• Select the right agent(s)
• Generate plan instructions
• Coordinate inter-agent workflows
• Collect and display results (including plots & summaries)

---

🧑‍💻 Developer Features

📁 Modular Agent System (DSPy)

Agents are implemented as dspy.Signature classes. Example:

class google_ads_analyzer_agent(dspy.Signature):
    goal = dspy.InputField(desc="User goal")
    dataset = dspy.InputField(desc="DataFrame")
    plan_instructions = dspy.InputField(desc="Instructions")
    code = dspy.OutputField(desc="Python code")
    summary = dspy.OutputField(desc="Analysis summary")

Add your own agent in minutes.

🔌 Built-in Dataset Connectors

• Ads: Google Ads, Meta, LinkedIn Ads
• CRM: HubSpot, Salesforce
• SQL: Postgres, MySQL, Oracle, DuckDB

Want more? Submit a request: Contact Us

---

🖼️ UI Feature Overview

Feature	Description
💬 Chat Interface	Ask questions and receive answers like a regular chat.
🧑‍💻 Code Editor	Inspect and edit generated code. Features include: AI-assisted edits, auto-fix for broken code.
📊 Analytics Dashboard (Enterprise)	Monitor usage, set limits, allocate credits, enforce roles & permissions.

---

🛠 Backend Highlights

• 🔧 Agent orchestration via DSPy
• 🧠 Model-agnostic LLM support
• 📈 Built-in chart formatter for best-guess visualization types
• 📂 Multi-agent workflows powered by centralized planner
• 🔄 Daily scheduled reports & auto-regeneration (enterprise-ready)

---

📅 Roadmap

🔜 Short-Term Goals

• [ ] Deep Analysis Mode (LLM equivalent of longform research)
• [ ] Multi-CSV / multi-sheet Excel analysis
• [ ] User-defined analytics agents via UI
• [ ] Improved code-editing and auto-debugging

🔭 Long-Term Vision

• Usability-First: Optimize UX through iteration and user feedback
• Community-Driven: Shaped by the global analyst community (follow us on Substack, LinkedIn)
• Open Collaboration: Build and share new agents, retrievers, and datasets

---

🧩 Contributing

We welcome contributions! You can:

• Add new agents
• Suggest UX improvements
• Contribute templates or datasets
• Submit bug reports or pull requests

📬 For collaboration or enterprise inquiries: https://www.autoanalyst.ai/contact

---

📄 License

Auto-Analyst is released under the MIT License — feel free to use, remix, and build on it.

🐦 Follow Us

• 🌐 Website
• 📰 Substack
• 💼 LinkedIn

---

Built with ❤️ by Firebird Technologies AI. Tech. Fire.