Metriq
Book a demo
Open source · BSL 1.1 · v0.4

Open Source
AI Notebook
for Data Science
and Analytics.

Open-Source Coding Agent Harness for Notebooks.

  • Built on JupyterLab — every extension, kernel, and shortcut already works.
  • Connect to your data, run the analysis, ship the result — end‑to‑end.
  • Claude, GPT, Gemini, or local models via Ollama. Bring your own keys.
  • Complete coding harness — skills, MCP tools, persistent memory.
Get started Talk to us
Powered by
Claude OpenAI Gemini Gemma DeepSeek Ollama
metriq · stripe-churn.ipynb kernel running
CHAT · 12:34
YOU
Why did churn jump in March? Pull from Stripe.
METRIQ Investigating
Querying Stripe, building a cohort, comparing March vs trailing 6‑mo baseline.
Connect stripe.events
Build cohort by signup month
Identify top 3 cancel reasons
Render heatmap + summary
› Ask Metriq… ⏎
NOTEBOOK · Python 3.12 +▶⏹
[1]
-- via stripe MCP
SELECT cohort_month, churn_rate
FROM stripe.events
GROUP BY 1 ORDER BY 1;
[2]
RETENTION HEATMAP — 12 cohorts × 6 months
March cohort churn 8.3% vs trailing baseline 5.1%. +63% MoM. Driver: failed renewals on annual plans.
● connected · stripe claude-opus-4.7 21 cells · 1.4s
live agent transcript · simplified fig.01
§ 01 — what you get

Three reasons analytics teams ship with Metriq instead of closed notebooks.

01

A real agent.
Not a chat box.

The agent plans, queries your warehouse, writes Python, executes cells, renders charts, and ships results — end to end. Every step appears as a real notebook cell you can run, edit, and rerun.

PlanExecuteShip
02

Runs on your laptop.

Five providers — Anthropic, OpenAI, Google, OpenRouter, Ollama. Air-gap deployment supported. Your data and your keys never leave your machine, full stop.

LocalAir-gapBYOK
03

Standard .ipynb.

A Metriq notebook is a Jupyter notebook. Open it in VS Code, Cursor, Colab, Kaggle, or any Jupyter tool. Every JupyterLab extension already works. Zero lock-in.

VS CodeCursorColab
§ 02 — how it works

From question
to result,
in one notebook.

No glue scripts. No hand-off between SQL editor, BI tool, and slide deck. The whole loop lives inside a single .ipynb.

01 / Connect

Bring your data.

Postgres, Snowflake, DuckDB on S3, Stripe, PostHog, your CSV. The agent uses MCP to discover schemas and run queries. Or skip the warehouse — DuckDB on parquet runs analytical workloads 20–50× cheaper than Snowflake at mid-market scale.

CONNECTIONS
Postgres prod
Stripe live
DuckDB / S3 parquet
PostHog events
+ Snowflake, BigQuery, MySQL, Salesforce… add
02 / Ask

Type a question, in plain English.

The agent generates a plan. You review and edit it. Then it executes — SQL queries, pandas code, matplotlib or plotly charts — all written into the notebook as cells you can run, modify, and rerun.

PLAN · 4 STEPS edit
  1. 01 Discover schema for stripe.subscriptions
  2. 02 Write SQL — cohort churn by month
  3. 03 Pivot to retention table in pandas
  4. 04 Render heatmap + write summary cell
03 / Ship

Send it where your team lives.

Weekly business review to Slack. PPTX deck to your COO. Streamlit dashboard for the team. The agent ships in the format the audience needs — and remembers how you like it for next time.

#analytics
Weekly review · 3 anomalies flagged
Mon 9am
Q1-board-review.pptx
14 slides · sent to COO
2.4 MB
churn-dashboard.streamlit.app
deployed · refreshes hourly
live
remembered for next week →
§ 03 — honest matrix

How we compare. Where they win, we say so.

Full compare
Metriq OPEN · LOCAL
Hex Deepnote Julius
Source available Yes · BSL 1.1 Closed Format only Closed
Self-host on your infra Yes Enterprise Roadmap No
Local LLMs (Ollama) All tiers No No No
Standard .ipynb Yes No (yaml) Convert No
JupyterLab extensions Yes No No No
Reactive DAG execution Roadmap Yes No No
§ 04 — talk to us

20 minutes. No demo theater.

We pull up your data in Metriq and answer one analytics question end to end. If we can't, we'll tell you why on the call.

Book a 20-min call Founder chat (15 min)
May 2026
‹‹ today›
MTWTFSS
TUE · MAY 12 · 2:30 PM PT
20-min intro with the Metriq team
§ 05 — get started

One command.
Everything included.

Free under $10M annual revenue. Source available under BSL 1.1. Use any model. Connect any data. Ship from your laptop.

GitHub 2.1k ★ Quickstart Discord 1.3k
~/projects · zsh 
$ pip install metriq
$ metriq notebook  # opens JupyterLab w/ agent

→  Connect data:  metriq connect postgres
→  Pick model:    metriq use claude-opus-4.7
→  Or local:     metriq use ollama/llama-3.3
python ≥ 3.11 · macos · linux · win/wsl ~3 min

Open Source AI Notebook
for Data Science
and Analytics.

Get started Talk to us