// the take
Infrahub is a source-of-truth platform for network/infrastructure data that layers git-style branching, diffing, and peer review on top of a Neo4j graph database, with Jinja2/Python generators to turn that data into configs and other artifacts. It's aimed at network and platform teams replacing NetBox/Nautobot-style tools or hand-rolled scripts with something that tracks who changed what and validates it before merge.
Applying git's branch/diff/review model directly to infrastructure data (not just to the code that generates it) is a real differentiator from static CMDBs like NetBox - you get a queryable history of schema and data changes, not just configs. Schema-first design backed by GraphQL gives you actual typing and relationship integrity instead of free-form key-value blobs. The generator/transform layer is explicitly built to be unit-tested and idempotent, which is the part that usually gets skipped in homegrown automation. It's not a single monolith either - there's a real SDK, an Ansible collection, a Nornir inventory plugin, and sync adapters for NetBox/Nautobot/IP Fabric, suggesting this is actually used in production rather than being a demo project.
The stack you need to run this is heavy for what is nominally a 'source of truth': Neo4j plus Git plus Prefect, versus a NetBox install that's just Postgres and Django. That's a real ops cost before you've modeled a single device. The README is almost entirely marketing copy - there's no actual GraphQL query example, no sample schema, no CLI snippet, so you can't tell what a day-one experience looks like without leaving the repo. Core production needs (HA, advanced RBAC, performance tuning) are explicitly carved out into the paid Enterprise edition, so the free tier's ceiling for a serious team isn't obvious from the repo alone. At 538 stars it's a small community relative to the incumbents (NetBox, Nautobot) it's positioned against, so expect fewer third-party schemas, Stack Overflow answers, or battle-tested deployment patterns than the tools it's competing with.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →