Why There Are So Many API Clients (and Why I Built yapi Anyway)

12/22/2025 · Source

Why There Are So Many API Clients (and Why I Built yapi Anyway)

If you search for "Postman alternative," you’ll find dozens of tools that all claim to fix the problem of being a great API client. That should have warned me that I was building something stupid.

However, it seems, API clients are not a solved problem. The way we write software keeps changing, and with LLMs entering the loop of our workflow, having tools which synergise the efforts of humans and LLMs is crucial.

This post reviews the current landscape of API clients, what each class of tool does well, where it breaks down, and why I ultimately built yapi instead of adopting a existing solution.

Postman and Insomnia: Interaction-First

The ol' classics. Everyone's favorite and least favorite.

Strengths

Good UX for exploration.
Easy to onboard non-engineers.

Limitations

The GUI is the source of truth.
State lives in synced workspaces and opaque JSON.
Automation is bolted on after the fact.
Not sexy.

curl, HTTPie,

xh: Command-First

Strengths

Ubiquitous and composable.
Easy to generate.
Ideal for quick debugging.

Limitations

No structure beyond the shell.
Chaining becomes fragile piping.
Assertions, reuse, and state require external glue.
Only kinda sexy.

Why this matters now LLMs are good at emitting commands. It would be a massive PITA to maintain a fleet of shell scripts plumbing all your curl integration tests together.

These tools are excellent primitives. They are not API testing systems.

Hurl: Test-First

To be fair, Hurl is really close conceptually to what I want. However, I didn't want a DSL for making requests, I wanted YAML with an LSP so anyone can onboard super fast and LLMs (and humans) can use the LSP to write correct code straight away.

Strengths

Declarative, readable, diffable.
CI-native by design.
Strong focus on assertions.

Limitations

HTTP-only, no gRPC, TCP, WebSockets.
'DS(m)L' weird hand-rolled domain specific markup language for defining requests. (Insert @theprimagen pronouncing it 'dismal')
Medium sexy.

Why this matters now Hurl is the closest existing tool to treating requests as code. But modern systems are rarely single-protocol or single-step. Once you need gRPC, raw sockets, or stateful workflows, you leave its design envelope.

Bruno

Strengths

Files are the source of truth.
Git-friendly and offline.
A principled response to Postman’s cloud model.
Pretty good CLI tool!

Limitations

Still too human-centric.

Why this matters now Bruno proves that “files over GUIs” was the right move. But it still assumes a developer is the one reading failures and deciding what to fix next.

Agents remain second-class users.

TUI Clients (Posting,

ATAC): Interaction Reimagined

Posting actually serialises requests in yaml very similar to yapi. It wouldn't be too hard to write a Posting adapter for yapi.

Strengths

Fast, keyboard-driven workflows.
Dramatically better than GUIs.
Pleasant for humans.

Limitations

The TUI is the product.
Files support the UI, not the other way around.
Automation is incidental.

Why this matters now If the interface is the abstraction, automation will always be downstream.

The Gap None of These Tools Close

All of the above tools were designed for a world where:

humans write the code,
automation is supplementary,
and APIs are mostly HTTP.

That world no longer exists.

Today:

LLMs generate and modify configs.
Regressions are caught automatically.
APIs span HTTP, gRPC, and raw protocols.
Text files are the universal interface.

No existing tool treats API interaction as a durable, executable artifact owned jointly by humans and agents.

That is the gap yapi exists to fill.

Why yapi Is Different (and Why It Had to Exist)

yapi makes a different set of tradeoffs, intentionally:

Files are the source of truth.
Workflows are declarative.
Assertions are first-class.
Output is structured and machine-readable.
yapi is extensible to any protocol.

This isn’t about being “better than Postman.” It’s about optimizing for a different future.