Skip to main content

​
Agentic Harness for Fast Moving Data Teams

SignalPilot is a Jupyter-native AI agentic harness that investigates data by connecting to your db warehouse, dbt lineage, query history, slack threads, notion, and Jira tickets — giving the AI institutional knowledge that coding assistants like ChatGPT and copilots can’t access. SignalPilot Agentic Harness: MCP Context Layer + Long-Running Agent Loop + Memory + Skills/Rules

Multi-Source Context

Aggregates organizational knowledgeDb warehouse, db warehouse, dbt lineage, database schemas, Slack discussions, Jira tickets, past investigations

Long Running Agent Loop

Plans, Executes, IteratesNot single-shot completions — continuous loop until task complete with analyst-in-the-loop

Multi-Session Memory

Remembers institutional knowledgePast hypotheses, validated assumptions, known data quirks, MCP subagent

Skills and Rules

Teach SignalPilot your domainCustom skills (reusable analysis patterns), coding standards (imports, style), business logic (revenue formulas)
🔒 Zero data retention • Read-only access • Local-first execution • SOC 2 in progress

​
How is SignalPilot different from ChatGPT or Copilot?

ChatGPT and IDE copilots are general-purpose code generators. SignalPilot is purpose-built for data investigation:

Full Org Context

ChatGPT/Copilots: Manual copy-paste or code files onlySignalPilot: Auto-connects to dbt, databases, Slack, Jira, query history via MCP

Multi-Step Investigations

ChatGPT/Copilots: Single-shot responses or single-file editsSignalPilot: Long-running loop with analyst-in-the-loop approval

Team Memory

ChatGPT/Copilots: No memory between sessions or notebooksSignalPilot: Remembers hypotheses, assumptions, data quirks

Custom Domain Logic

ChatGPT/Copilots: Generic suggestionsSignalPilot: Your team’s skills, rules, and business logic
Why “harness”? SignalPilot provides the infrastructure that lets AI work on complex data investigations while keeping you in control.

See Full Comparison with Real Examples

SignalPilot vs ChatGPT vs Copilot — Head-to-head workflows + when to use each

​
One Minute Installer

Prerequisites: macOS, Linux, or Windows (WSL) • uv • Internet connection Don’t have uv? Install it first (takes seconds): 
curl -LsSf https://astral.sh/uv/install.sh | sh
Then install SignalPilot: 
uvx signalpilot
What happens: 
✓ Creating workspace at ~/SignalPilotHome
✓ Installing isolated Python 3.12 + Jupyter Lab + SignalPilot extension
✓ Installing data packages (pandas, numpy, matplotlib, plotly...)
✓ Optimizing Jupyter cache for fast startup
⚡ Setup complete in 1m 10s

→ Launching Jupyter Lab at http://localhost:8888
Option 1: Using uvx (recommended)
uvx signalpilot home
Option 2: Manual activation
cd ~/SignalPilotHome
source .venv/bin/activate && jupyter lab

Full Installation Guide

Alternative methods, manual installation (pip, conda, uv), troubleshooting

​
Get Started in 3 Steps

Launch SignalPilot

uvx signalpilot home
Opens Jupyter Lab in your SignalPilot workspace at http://localhost:8888
uvx signalpilot home is shorthand for uvx signalpilot lab --home
When creating notebooks, always use the default Python 3 kernel. SignalPilot uses this kernel and may not work with other kernels due to missing system variables.
Run the installer first:
uvx signalpilot
Takes ~2 minutes to set up everything

Connect your data (optional)

Start immediately with local files:
  • Type @ in chat to mention any CSV, Excel, or data file from the data/ folder
  • Home workspace: ~/SignalPilotHome/data/
  • Project folder: ./data/ in current working project directory
Add more folders: Use the File Scanner tool in the left sidebar to index additional directories.Connect databases: PostgreSQL, Snowflake, BigQuery for warehouse access.

Database Setup

Connect PostgreSQL, Snowflake, BigQuery

Context & @Mentions

Reference local files, dataframes, and databases

Ask your first question

“Show me the top 10 customers by revenue this month”
SignalPilot generates a plan → you approve → it executes

5-Minute Quickstart

Complete walkthrough with real example

​
Learn More

How It Works

4 foundational systems explainedMCP context layer, long-running loop, memory & hooks, skills & rules

Why SignalPilot?

vs ChatGPT vs IDE CopilotsFull comparison + real-world investigation example

Security & Privacy

Read-only, local-first, zero retentionSOC 2 in progress, GDPR compliant, audit trail

All Integrations

Connect your data stackPostgreSQL, Snowflake, dbt, Slack, Jira, and more

​
Resources

Quickstart Tutorial

5-minute investigation walkthrough

CLI Reference

Commands and troubleshooting

Configuration

Customize behavior and rules

Discord Community

Get help and share use cases

GitHub

Report issues and contribute

Documentation

Full concepts and guides