OpsGraph

Ask PostgreSQL a question. Check the records behind the answer.

A private, self-hosted workspace for investigating missing records, failed payments and stuck jobs. Ask a question, inspect the SQL and saved results, then continue in the same conversation. Export a readable incident report when another engineer needs the evidence.

Get started · First investigation · Stop and return · Setup guide · Security

Live-tested with Anthropic and synthetic Supabase payment data. An incident question, a conversation without a database query, and a follow-up using fresh evidence stay together. Click the image for the full page.

Get started

You need Git, Python 3.9 or newer, and a browser. The installer prepares OpsGraph's own Python environment and asks before downloading software.

On macOS or Linux, open Terminal and run:

git clone https://github.com/Arittra-Bag/opsgraph.git
cd opsgraph
python3 Start.py --no-code

On Windows, use PowerShell and replace the last line with py Start.py --no-code. Already cloned? Run the last command from your opsgraph folder.

Choose your path:

Choice	What you need	What happens
Try with practice data	PostgreSQL 15-18 installed, or local Docker Desktop running	Creates a separate database with invented orders and payments and a read-only login
Connect my database	A PostgreSQL connection string with a restricted read-only login	Guides you through your hosting provider's connection settings

Then choose a model service: your own hosted API key, or an already running local model. OpsGraph does not install PostgreSQL or download model files. The setup asks before saving and opens your private workspace in the browser. Passwords and keys stay on your backend and are hidden while you type them.

Need help getting a database login? Choose the setup help option, or skip the connection and use the administrator guide in Sources later. Never use your database owner's login. Step-by-step setup.

First investigation

1. Sources: choose the tables OpsGraph may read. Check the connection and approve the small test read. Practice setup can do this after your approval.
2. Settings: save your model choice, then click Test actual model connection.
3. Investigations: ask a specific question and open the cited evidence.

For the included practice database, try:

Count payments by processor, status and error_code. Which processor has failed payments, how many are there, and what error is recorded? Cite the evidence.

Continue with follow-ups in the same investigation. Earlier results stay saved, and retries keep their own attempts. Product questions do not read your database. Use Create report to preview and download Markdown. Choose whether to include SQL and source records, and review the content before sharing.

Stop and return

Press Ctrl+C in the launch terminal. Closing the browser does not stop OpsGraph. Return with the same command from the same checkout:

python3 Start.py --no-code

Your settings and history stay saved. To review setup again:

python3 Start.py --no-code --configure

To update, stop OpsGraph, back up your workspace, then run git pull --ff-only and the same launch command. It prepares the updated application when needed. Maintenance and recovery.

Connections and privacy

Use local or hosted PostgreSQL, including Supabase, Neon, AWS RDS, Google Cloud SQL, Azure and DigitalOcean. Hosting choices supply setup guidance, not provider certification. Remote connections verify the server's certificate and hostname with sslmode=verify-full. Connection guide.

Use Ollama, LM Studio, vLLM, Anthropic, OpenRouter, OpenAI, Groq, Together, Mistral, or a compatible custom endpoint. You choose and test the model. OpsGraph does not automatically discover models or switch providers after a failure.

External model processing requires explicit permission. Approved schema, questions and bounded query results may reach that provider. Local inference is available when you supply a local model. Reports and backups can contain sensitive data. Security and data handling.

Supported boundary

One trusted operator, PostgreSQL, and read-only investigations. Keep the service on your computer or a private network. The model proposes queries, while application checks and PostgreSQL permissions decide what can run.

Each data attempt allows at most three queries, 100 captured rows per query and five seconds per query, with stricter playbook limits where applicable. Changing an approved schema invalidates its previous checks. A citation or matching hash does not prove an answer correct. Review evidence and supply business definitions.

No database writes, automatic fixes, team accounts or continuous monitoring. Validation and platform limits · Production boundary.

Other setup options

• Quick or Advanced: python3 Start.py uses the original guided workspace. Installation options.
• Docker: runs the application container. Your database and model remain separate. Container setup.
• Offline bundles: use an artifact matching its published version and platform. Bundle guide.

Development checks

Keep changes scoped and never commit credentials, state databases or private exports. Run from the repository root:

uv run ruff check src tests scripts Start.py
uv run ruff format --check src tests scripts Start.py
uv run python -m pytest -q
node --test tests/frontend/*.test.cjs
uv build --build-constraints requirements-build.lock --require-hashes
uv run python scripts/wheel_smoke.py

The wheel smoke test expects one wheel in dist. Frontend tests need Node.js. Read the connector contract and threat model before changing execution or access controls.

Report a bug with your version, OS and sanitized error. Report security issues privately. Changelog · Backup and restore · Upgrade notes.

Licensed under Apache-2.0. Dependencies retain their own licenses.