Test GraphQL APIs as files in your own repo.

Both jobs, one mechanism. A `graphql` block on a request node sends the query, and the same node carries assertions, capture, retry, hooks and auth, so a single .aether file is the request and the test of it. The file is plain YAML in your own Git repo, the graph canvas and the code editor are two renderings of it, and one local engine runs it in the desktop app, the CLI, and CI. There is no account and it works offline against anything you can reach.

It fails the node, which is the behaviour most tools get wrong by default. GraphQL servers return HTTP 200 for execution errors, so a status assertion alone passes on a response whose errors array says the query broke. If the response carries a non-empty errors array, the node fails with the first error's message. When you want to assert on the errors themselves, set allowErrors: true in the block and read response.body.errors like any other value.

Yes, and this is where a flow format earns its keep. GraphQL is a block on the ordinary request node, so a flow can log in over REST, capture the token, run a GraphQL mutation with it, then check a side effect over gRPC or wait on a WebSocket message, with conditions, loops and parallel fan-out between the steps. Nothing changes shape between protocols: every step stores { status, headers, body } to assert on and capture from.

No, and this page will not pretend otherwise. The graphql block is written as YAML in the code editor: the query is a YAML block scalar, variables are a map, and there is no structured panel, no introspection-driven autocomplete, and no schema browser. If a point-and-click query builder is the feature you are choosing a tool for, Postman, Insomnia and Bruno all have one and Bandura today does not. What Bandura adds instead is everything around the query: assertions, capture, chaining, a debugger, and CI.

Partially, and the edges are named rather than fudged. A Bruno request with a body:graphql block imports as a real graphql request node, query and variables intact. Postman and Insomnia GraphQL bodies are currently skipped rather than half-imported, because a silently mangled query is worse than a named gap: the rest of the collection comes across and the GraphQL requests are listed as skipped in the review step.

The same .aether file, headless. The CLI installs with npm install -g @bandura/cli and ships as a compiled binary with the engine inside, so bandura run flows/ executes exactly what the desktop app executes: deterministic exit codes, JUnit XML for the dashboard, GitHub annotations on the pull request, a concurrency flag, and run --repeat with p95/p99 thresholds when a latency gate is wanted.

The download is the complete app, free indefinitely for personal projects, learning, and open source. Commercial use takes a one-time per-seat licence after a 30-day evaluation that disables nothing; the price is announced when checkout opens at launch. It covers every release built in the following 12 months, with a perpetual fallback: when the window ends, the build you have keeps working, permanently, and nothing auto-renews.

As a variables map in the graphql block, so the query stays a clean document instead of a string you concatenate values into. Everything in the map interpolates with ${{ }} exactly as a URL or header does: a captured id from the login step, an env.* value, or a dynamic helper such as $uuid(). Bandura assembles the { query, variables, operationName } JSON body for you, and captures read the result under response.body.data, so the value one query returns feeds the next one's variables.