Turn your database into decision intelligence.
Ask a business question. Insyte turns it into validated, read-only SQL, masks sensitive data, logs the execution, and returns charts and insights—without exposing your database credentials to AI.
pip install insyte
Monthly revenue analysis
The database URL remains in a local permissioned file and is never passed to the model.
Every query travels through the guardrail.
Insyte does not give an AI agent unrestricted database access. It converts intent into a constrained request, validates the SQL, enforces limits, masks sensitive fields, and records the execution before returning results.
AI models never receive your database credentials.
Your connection URL stays on your machine in a permissioned local file. It is not written into config, logs, prompts, or MCP payloads.
local file permission · 0600Dangerous SQL is blocked before execution.
Every generated query is inspected against Insyte's safety rules. Unsupported statements are rejected rather than executed and cleaned up afterward.
validate → limit → mask → auditFrom quick questions to full investigations.
Explore live data in Studio, work from the terminal, or call the same safe capabilities from Claude Code and Codex through MCP.
Use Insyte where your work already happens.
Move from a browser workspace to terminal workflows or your AI coding agent without changing the safety model underneath.
See movement, not just totals.
Break down metrics by segment, compare periods, and inspect the exact SQL behind every result.
Ask “why did it change?”
Run a safe multi-step analysis instead of stopping at the first chart.
Turn results into an analyst-ready brief.
Generate summaries, evidence, counter-evidence, forecasts, and recommendations from already aggregated results.
Revenue performance brief
Generated from grounded resultsFollow-up questions stay connected.
Studio remembers the active metric, dimension, and period, so natural follow-ups resolve correctly.
total_revenuecitylast_monthsame for previous monthThree commands to your first answer.
Run Insyte inside a virtual environment, connect a read-only PostgreSQL role, and let the guided setup scan your schema and generate local metadata.
Create a virtual environment
Keep the installation isolated from the rest of your Python tooling.
python -m venv .venv && source .venv/bin/activateInstall Insyte
Get the Python package from PyPI.
pip install insyteRun guided setup
Provide a read-only connection URL, choose an AI tool, and let Insyte scan the schema and wire up MCP.
insyte initOpen Studio
Launch the localhost browser workspace and start asking questions.
insyte studioEverything you need, one command away.
Your project data lives under ~/.insyte/projects/<name>/. The database URL is loaded only when required and is not written to the main config file.
Ask better questions. Keep control of your data.
Install Insyte, connect a read-only PostgreSQL role, and explore your data through Studio, CLI, Claude Code, or Codex.