OpsGraph BETA
Release View on GitHub
Public-validation beta · v0.1.0b1 Apache-2.0

Ask PostgreSQL.
Inspect the evidence.

OpsGraph is a self-hosted investigation workspace for one operator. Your model plans bounded, read-only queries. Application policy controls their execution.

Try the beta See how it works

Built for evaluation on an authorized, non-sensitive PostgreSQL test database. Not production-ready.

Self-hosted Read-only PostgreSQL External inference off by default One operator
OpsGraph logo Policy-controlled investigation
Recorded execution inspectable
  1. 01
    Scope checkedApproved tables + playbook
  2. 02
    Query plannedModel proposes bounded SQL
  3. 03
    Policy enforcedRead-only limits before execution
  4. 04
    Evidence retainedSQL, rows, source + limits
SELECT … LIMIT 100 5s timeout
Query
Evidence
Read-only
MODEL PROPOSES•APP VALIDATES•POSTGRESQL STAYS READ-ONLY•EVIDENCE REMAINS INSPECTABLE•OPERATOR DECIDES
Why OpsGraph

Evidence before certainty.

A citation does not make a conclusion true. OpsGraph keeps the work inspectable so you can review what was actually queried and captured before you act.

See what happened

Open referenced evidence to inspect the query, source, collection time, captured rows and limits behind a finding.

Recorded execution

Keep access bounded

Choose explicit schema-qualified tables and use a dedicated read-only role. The model cannot grant itself access or make database changes.

Policy before execution

Keep the investigation

Reopen history, export evidence, retry failed work with fresh queries, or ask a linked follow-up while earlier evidence remains unchanged.

Saved local history
The boundary is part of the product

The model does not get the keys.

The model can plan. OpsGraph validates and executes inside explicit application policy. Database permissions remain the final authority.

SELECT only Approved tables Role checks Read-only transactions

Current beta defaults allow up to three queries, 100 captured rows per query, and five seconds per query. Playbooks may be stricter.

execution_boundary.policy ENFORCED
database_actionSELECTallowed
table_scopeapproved onlychecked
query_budget3 maxbounded
row_capture100 / querybounded
statement_timeout5 secondsbounded
INSERT / UPDATE / DELETEdeniedblocked
Application policy controls execution — not the model.
One investigation, four inspectable stages

Ask. Bound. Capture. Review.

The workflow is intentionally narrow. OpsGraph does not silently invent sample answers, download models, or remediate your database.

01

Ask a concrete question

Pick a source and playbook. Start with a question whose columns and meanings you actually know.

02

Plan inside a scope

Your configured model proposes bounded queries against tables you explicitly approved.

03

Execute read-only

AST validation, role checks, approved-table scope and read-only transactions gate execution.

04

Inspect the evidence

Compare the answer with captured rows and SQL. Review missing evidence and limitations before acting.

Real beta workspace

The answer stays attached to its work.

A saved investigation can be reopened later. Findings, cited evidence, captured records and limitations remain inspectable, while follow-ups collect fresh evidence.

SQLinspectable Rowscaptured Limitsrecorded Follow-upfresh queries
opsgraph / investigation
OpsGraph beta investigation workspace showing recorded execution, findings, evidence and a follow-up form
Saved investigation from beta validation · test data only
Bring your model

Local when you want it. Explicit when it leaves the machine.

Use Ollama locally, configure another OpenAI-compatible endpoint, or connect Anthropic with explicit external-data permission. OpsGraph does not download model weights for you.

LOCAL

Ollama

Tested local path. Keep model inference on your own machine when that fits your workflow.

API

OpenAI-compatible

Point OpsGraph at an endpoint you operate or trust. Provider access stays explicit.

API

Anthropic

External inference is opt-in. Review data handling before connecting source records.

Get started

Choose the path that matches what you are doing.

If you are evaluating OpsGraph, the offline bundle is the shortest path. If you want to modify or contribute, use the source checkout with uv.

Evaluating?Use the beta bundle.CPython 3.11 is required; the bundle carries the app and locked dependencies, not the interpreter.
Contributing?Use the source path.Python 3.11–3.13 + uv gives you the editable environment, tests and build workflow.
Best for trying the beta without a development checkout.
# Download the matching v0.1.0b1 bundle + .sha256
# macOS / Linux
python3.11 -I Install.py install
python3.11 -I Install.py launch

# Windows
py -3.11 -I Install.py install
py -3.11 -I Install.py launch
Open release assets ↗
What “beta” means here

Useful enough to evaluate. Narrow enough to be honest about its limits.

Beta 1 is a prerelease for public validation. Native installer lifecycles are checked on macOS, Ubuntu and Windows, but validation depth differs by platform. The Docker image has configuration and health checks; a complete container investigation workflow is not yet claimed as verified.

Read the support matrix Report an issue
Public validation is open

Try it on a non-sensitive test database. Challenge the evidence model.

The most useful feedback is technical: execution boundaries, evidence quality, installation friction and operator workflow.

Explore the repository Get Beta 1