The first useful thing Hasura built was not Hasura. It was a food-delivery prototype, conceived by Tanmai Gopal and Rajoshi Ghosh around a marketplace for home-cooked meals. The app itself came together. Then reality knocked: recruiting cooks, moving dinners across a city and making the unit economics behave were the difficult bits. Software could not keep the dal warm.
But the founders noticed something hiding in the wreckage. They had postponed building the technology because they expected it to be quick, then found themselves wanting a reusable way to make every future application quicker still. Consulting projects reinforced the pattern. Again and again, teams needed the same unglamorous machinery: database access, authentication-aware permissions, events and APIs. The proposed food business had a logistics problem. Its builders had discovered a software problem that repeated.
The product is the missing plumbing
Hasura's original sharp trick was easy to demonstrate. Connect a PostgreSQL database and the GraphQL Engine generates a queryable, real-time API. Define roles and row- or column-level permissions in metadata. Add event triggers and remote schemas when the application needs custom behavior. A developer can skip a pile of hand-written schemas, resolvers and authorization checks, yet still work with the existing database.
That does not abolish backend engineering. It removes a repetitive slice of it. Teams still design data, authentication and business rules; they still own the parts that make their product peculiar. Hasura handles the common path between a client and its data, compiling a GraphQL request into efficient work for the underlying source. The distinction explains both the appeal and the limits.
APIs · code
metadata + policy
services · AI
“We charge for constructing the bridge, not the toll for using it.”Hasura's description of DDN active-model pricing
The open-source engine gave Hasura an efficient route into companies. An individual developer could get the immediate win for free. If that API became production infrastructure, the organization acquired less playful concerns: availability, caching, tracing, rate limits, compliance, premium connectors, support and private deployment. Hasura Cloud operates the system; Enterprise Edition sells those controls to companies that need to self-host. The community product supplies adoption. Operational anxiety supplies the budget.
What broke first - and what widened
The early false start was the food marketplace: the hard part was operations, not code. The next constraint was subtler. A Postgres-first GraphQL engine is a fine wedge, but enterprise data does not live in one neat Postgres database. It sits in SQL Server and Snowflake, behind REST services, inside old systems, across clouds and under different teams. Custom business logic also strained against a product best known for instant CRUD. Hasura's own DDN alpha notes acknowledged complaints about metadata authoring, deployment time, cold starts and the burden of hosting simple functions elsewhere.
DDN, generally available in August 2024, is the answer. It treats Hasura as a metadata-driven data access layer rather than merely a GraphQL generator. Connectors translate standardized requests into native queries for sources ranging from PostgreSQL and MongoDB to Oracle, Elasticsearch and Stripe. A build system creates preview API deployments and rollbacks. Federated metadata allows domain teams to contribute to one graph without dumping all ownership into a central group. The execution engine was rearchitected, with Rust taking a larger role in the next-generation runtime after the original engine's deep association with Haskell.
Pricing moved with the product. Hasura Cloud once advertised a $99 monthly Standard plan, then introduced a Professional tier starting at $1.50 an hour. DDN's beta proposed charging per “active model” - a model receiving at least 1,000 hits per month - rather than metering every request. At announcement, Hasura listed free access for one developer, Base at $5 per active model monthly, Advanced at $30, and Private DDN from $1,000 per availability zone monthly. Those figures describe the launch model, not a permanent quote; infrastructure pricing has a habit of becoming historical material while the tab is still open.
Customers buy speed, then boundaries
The customer roster spans startups and large organizations. Optum used Hasura while moving a clinical platform from concept to production in 100 days, without pretending every piece of data would sit in the new platform. Nutrien connected an Aurora PostgreSQL database while building a web, iOS and Android hub across a business assembled from multiple companies and more than 2,000 retail locations. Public customer material also names Philips Healthcare, HMH, Fieldguide, Pipe and General Mills.
They are not all buying GraphQL as a fashion accessory. They are buying controlled access across organizational seams. Product teams want to ship screens without waiting for a custom endpoint. Platform teams want one place to express permissions and observe requests. Architects want to modernize a legacy estate without migrating every byte before users see value. Hasura competes here with Apollo's federation tooling, AWS AppSync, PostGraphile, backend platforms such as Supabase and Firebase, API products such as Kong and MuleSoft, and the eternal alternative: a team writing services by hand.
The company mirrors the distributed systems it sells. Hasura describes a globally distributed team with office options in San Francisco and Bangalore, flexible schedules and a monthly second Friday off. Its published values are User First, One Team and Win or Learn - the last of which fits a company whose origin story depends on treating a discarded plan as usable evidence. LinkedIn places the team in the broad 51-to-200-employee band, but Hasura does not publish a precise current headcount. The open-source culture is more visible than the org chart: public issues, community discussions, connector specifications and regular releases let customers inspect the machinery and argue with its makers.
The AI chapter is really a data chapter
In June 2025, Hasura announced PromptQL in beta, calling it the spiritual successor to GraphQL for the AI era. The pitch shifts from generated CRUD APIs to generated programs: a user asks a natural-language question or requests a task, and the system plans work against governed data and business logic. GraphQL Engine and DDN remain maintained products, while PromptQL widens the audience from application developers consuming data to people and agents trying to use it.
This is a credible adjacency because AI applications inherit the old problem with extra ways to fail. A language model can compose a plausible sentence; it still needs accurate context, permission-aware retrieval, repeatable business logic and a trace of what happened. Hasura's expertise is not model training. It is the controlled path from a request to operational data. PromptQL is a bet that this path becomes more valuable when the requester speaks English instead of GraphQL.
If nearly every request is a bespoke workflow, generated data access may save little and add another critical layer. Event triggers can create operational coupling if teams treat them like a guaranteed message bus without designing for overload and retries. A connector is only as useful as its maturity. And organizations unwilling to place authorization metadata outside existing services may dislike Hasura's center of gravity. The tool works when its conventions cover a large, stable portion of the workload.
The piece worth stealing
Hasura's most portable lesson arrived before GraphQL. The founders did not conduct a grand search for “developer infrastructure.” They attempted a business, noticed which work repeated, and separated the reusable machine from the customer-specific mess. Consulting gave them more repetitions. Open source made the machine cheap to test and easy to distribute. Only after developers pulled it into important systems did the company package the expensive organizational needs around it.
A builder can copy the sequence without copying the product. First, list the components your team rebuilds across projects. Pick one with a result users can experience in minutes. Make the boundary explicit: what disappears, what remains theirs, and what breaks under unusual logic. Let the early product stay narrow long enough to become dependable. Then follow real adoption into adjacent sources, controls and deployment modes.
The method would not have worked if Hasura's internal tool solved only its founders' peculiar stack, if GraphQL had not given developers a recognizable interface, or if the generated layer could not preserve fine-grained permissions. Nor would the open-source funnel work without a production product worth buying. “Free code, paid fear” sounds cynical; the less cute version is that companies pay someone to own reliability once a convenience becomes infrastructure.
Hasura began by asking why an application could not be built in a week. More than a decade later, the wording has changed - API platform, supergraph, data delivery network, AI data access - but the irritation is intact. Useful companies often look like that: one stubborn complaint, widened carefully enough to survive several technology seasons.