Launching today
OpsGraph
Evidence-first agentic investigations for PostgreSQL
1 follower
Evidence-first agentic investigations for PostgreSQL
1 follower
OpsGraph is a self-hosted, public-validation alpha for evidence-first PostgreSQL investigations. A model may propose bounded read-only queries; OpsGraph enforces schema and table scope, timeouts, row limits, policy gates, source-owned evidence bindings, citations, and a local audit chain. Start with the bundled synthetic workspace. Use only synthetic data or non-production replicas for now.
Hi, I’m Arittra, and I’ve been building OpsGraph.
The idea started from a fairly simple constraint: an AI system investigating operational data probably shouldn’t be given an unrestricted database connection and a vague prompt.
So for this alpha, I’ve intentionally kept the scope narrow:
a separately provisioned, read-only PostgreSQL role
AST-validated, bounded `SELECT` queries executed in read-only transactions
source-owned evidence bindings, so the model can’t simply declare its own evidence coverage
evidence hashes and a local tamper-evident audit chain
local-first operation, with cloud-model egress only when explicitly enabled
This is still a public-validation alpha, not something intended for production or customer data.
I’d especially appreciate feedback from platform, SRE, and data teams: what would you need to see before you’d feel comfortable trusting a self-hosted investigation workspace against a non-production replica?