AsyncAPI 3.0 specification generation for Rust WebSockets and async protocols
Generate AsyncAPI documentation directly from your Rust code using procedural macros. Similar to how utoipa generates OpenAPI specs for REST APIs, asyncapi-rust generates AsyncAPI specs for WebSocket and other async protocols.
- Features
- Migrating from 0.3.x
- Migrating from 0.2.x
- Quick Start
- Examples
- Motivation
- Comparison: Manual vs Generated
- Supported Frameworks
- Binary Protocol Support
- DateTime Support (Chrono)
- Generating Specification Files
- Documentation
- Roadmap
- Contributing
- License
- Statement on AI/LLM Usage
- Acknowledgments
- 🦀 Code-first: Generate specs from Rust types, not YAML
- ⚡ Compile-time: Zero runtime cost, all generation at build time
- 🔒 Type-safe: Compile errors if documentation drifts from code
- 🎯 Familiar: Follows patterns from
utoipa,serde, andclap - 🌐 Framework agnostic: Works with actix-ws, axum, or any serde-compatible types
- 📦 Binary protocols: Support for mixed text/binary WebSocket messages (Arrow IPC, Protobuf, etc.)
Breaking change in 0.4.0: all map-typed fields on model structs (servers, channels, operations, components.messages, components.schemas, properties, etc.) changed from std::collections::HashMap to indexmap::IndexMap. This makes generated spec output byte-stable across builds — previously HashMap's random iteration order caused churn in downstream TypeScript codegen and other consumers even when the API hadn't changed.
If you construct model structs directly (e.g. in tests or a custom spec builder), replace HashMap::new() with IndexMap::new(). indexmap is re-exported from asyncapi_rust so you don't need to add it to your own Cargo.toml:
use asyncapi_rust::indexmap::IndexMap;
let mut servers = IndexMap::new();
servers.insert("production".to_string(), my_server);Code that only reads from these fields (iterating, .get(), .contains_key()) compiles unchanged — IndexMap has the same API as HashMap for read operations.
Breaking change in 0.3.0: message names in components.messages and asyncapi_message_names() now derive from the Rust variant identifier, not the serde rename string.
| Before (0.2.x) | After (0.3.0) |
|---|---|
messages.get("user.join") |
messages.get("UserJoin") |
asyncapi_message_names() → ["user.join", …] |
asyncapi_message_names() → ["UserJoin", …] |
The serde rename string remains the wire discriminant inside the payload schema — wire format is unchanged.
Other 0.3.0 additions:
#[asyncapi(message_name = "CustomName")]per-variant override for disambiguation- Runtime collision detection: two enums sharing a variant identifier in the same document panic with a clear error instead of silently overwriting
Schema::Anyhandlesserde_json::Valuefields without panicking- AsyncAPI 3.0–compliant
Parameterobject (removedschema, addeddefault,enum_values,examples,location) - Shared
$defsare hoisted tocomponents.schemas; message payloads reference them via$ref: "#/components/schemas/X"
Add to your Cargo.toml:
[dependencies]
asyncapi-rust = "0.4"
serde = { version = "1.0", features = ["derive"] }
schemars = { version = "1.1", features = ["derive"] }
# Optional: for chrono datetime support in schemas
chrono = { version = "0.4", features = ["serde"] }
schemars = { version = "1.1", features = ["derive", "chrono04"] }Define your WebSocket messages:
use asyncapi_rust::{schemars::JsonSchema, ToAsyncApiMessage};
use serde::{Deserialize, Serialize};
/// WebSocket messages for a chat application
#[derive(Serialize, Deserialize, JsonSchema, ToAsyncApiMessage)]
#[serde(tag = "type")]
pub enum ChatMessage {
/// User joins a chat room
#[serde(rename = "user.join")]
UserJoin { username: String, room: String },
/// Send a chat message
#[serde(rename = "chat.message")]
Chat { username: String, room: String, text: String },
}
fn main() {
// Get message names — returns Rust variant identifiers, not serde rename strings
let names = ChatMessage::asyncapi_message_names();
println!("Messages: {:?}", names); // ["UserJoin", "Chat"]
// Generate messages with JSON schemas
let messages = ChatMessage::asyncapi_messages();
// Each message includes:
// - name and title
// - contentType: "application/json"
// - payload: Full JSON Schema from schemars
let json = serde_json::to_string_pretty(&messages).unwrap();
println!("{}", json);
}Combine message types into complete specifications using #[asyncapi_messages(...)]:
use asyncapi_rust::{AsyncApi, ToAsyncApiMessage, schemars::JsonSchema};
use serde::{Deserialize, Serialize};
// Define your message types
#[derive(Serialize, Deserialize, JsonSchema, ToAsyncApiMessage)]
#[serde(tag = "type")]
pub enum ChatMessage {
#[serde(rename = "user.join")]
UserJoin { username: String, room: String },
#[serde(rename = "chat.message")]
Chat { username: String, text: String },
}
// Reference message types in your API spec
#[derive(AsyncApi)]
#[asyncapi(title = "Chat API", version = "1.0.0")]
#[asyncapi_messages(ChatMessage)] // Automatically includes all messages
struct ChatApi;
fn main() {
let spec = ChatApi::asyncapi_spec();
// spec.components.messages now contains all ChatMessage variants
// with full JSON schemas
}The #[asyncapi_messages(...)] attribute automatically populates the components/messages section with:
- All message definitions from referenced types
- Complete JSON schemas generated from Rust types
- Message metadata (name, summary, description, content-type)
Define dynamic server paths and channel parameters for WebSocket connections:
use asyncapi_rust::AsyncApi;
#[derive(AsyncApi)]
#[asyncapi(title = "User WebSocket API", version = "1.0.0")]
#[asyncapi_server(
name = "production",
host = "api.enlightenhq.com",
protocol = "wss",
pathname = "/api/ws/{userId}",
variable(
name = "userId",
description = "Authenticated user ID",
examples = ["12", "13"]
)
)]
#[asyncapi_channel(
name = "rtMessaging",
address = "/api/ws/{userId}",
parameter(
name = "userId",
description = "User ID for this WebSocket connection",
examples = ["42", "100"]
)
)]
struct UserApi;Server variables define placeholders in server URLs with:
name: Variable name (required)description: Human-readable descriptionexamples: Example values for documentationdefault: Default value if not providedenum_values: Restricted set of allowed values
Channel parameters define path parameters with:
name: Parameter name (required)description: Human-readable descriptiondefault: Default value if not providedenum_values: Restricted set of allowed values (e.g.,["v1", "v2"])examples: Example values for documentation (e.g.,["42", "100"])location: Runtime expression for the parameter's location
By default, message names in components.messages and asyncapi_message_names() are the Rust variant identifiers (UserJoin, Chat), not the serde rename strings ("user.join", "chat.message"). The serde rename is preserved as the wire discriminant inside the payload schema.
If two ToAsyncApiMessage enums in the same AsyncAPI document share a variant identifier, the runtime detects the collision and panics with a clear message. Use #[asyncapi(message_name = "…")] to disambiguate:
#[derive(Serialize, Deserialize, JsonSchema, ToAsyncApiMessage)]
#[serde(tag = "message")]
pub enum Operation {
#[serde(rename = "get-info")]
GetInfo { project_id: i64 },
}
#[derive(Serialize, Deserialize, JsonSchema, ToAsyncApiMessage)]
#[serde(tag = "message")]
pub enum OperationResponse {
// Same wire discriminant as Operation::GetInfo, but a distinct message name
#[serde(rename = "get-info")]
#[asyncapi(message_name = "GetInfoResponse")]
GetInfo { id: i64, label: String },
}Both messages appear in components.messages under distinct keys (GetInfo and GetInfoResponse), with "get-info" as the wire value in both payload schemas.
#[asyncapi(message_name = "…")] attributes:
message_name = "CustomName": Override the default (variant ident) for a single variant- An empty serde rename (
#[serde(rename = "")]) automatically falls back to the variant identifier — no override needed
See working examples in the examples/ directory:
simple.rs- Basic message types with schema generationchat_api.rs- Complete AsyncAPI 3.0 specification with server, channels, and operationsmessage_integration.rs- Automatic message integration with#[asyncapi_messages(...)]server_variables.rs- Server variables and channel parameters for dynamic pathsasyncapi_derive.rs- Using#[derive(AsyncApi)]for specsfull_asyncapi_derive.rs- Complete spec with servers, channels, operationsgenerate_spec_file.rs- Generating specification filesactix_websocket.rs- Real-world actix-web + actix-ws integrationaxum_websocket.rs- Real-world axum WebSocket integrationframework_integration_guide.rs- Comprehensive framework integration guide
Run any example:
cargo run --example simple
cargo run --example message_integration
cargo run --example server_variablesManually maintaining AsyncAPI specifications is error-prone and time-consuming:
- ❌ Type changes in Rust require manual YAML updates
- ❌ No compile-time validation of documentation accuracy
- ❌ Easy for docs to drift from implementation
- ❌ Repetitive work defining the same types twice
asyncapi-rust solves this by generating AsyncAPI specs directly from your Rust types, providing a single source of truth with compile-time guarantees.
Before (Manual YAML):
# asyncapi.yaml - must keep in sync manually!
components:
messages:
SendMessage:
payload:
type: object
properties:
type: { type: string, const: SendMessage }
room: { type: string }
text: { type: string }After (Generated from Rust):
/// Send a chat message
#[derive(Serialize, Deserialize, ToAsyncApiMessage)]
#[serde(tag = "type", rename = "SendMessage")]
pub struct SendMessage {
pub room: String,
pub text: String,
}
// AsyncAPI YAML generated automatically at compile time!- ✅ actix-ws - Full integration with actix-web WebSocket handlers
- ✅ axum - Integration with axum WebSocket routes
- 🔄 Framework-agnostic - Works with any serde-compatible message types
Document binary WebSocket messages (Arrow IPC, Protobuf, MessagePack):
/// Binary data stream
#[derive(ToAsyncApiMessage)]
#[asyncapi(
content_type = "application/octet-stream",
triggers_binary,
description = "Raw binary data payload",
)]
pub struct BinaryData;asyncapi-rust uses schemars 1.1 with full support for chrono datetime types:
use asyncapi_rust::{schemars::JsonSchema, ToAsyncApiMessage};
use chrono::{DateTime, NaiveDateTime, Utc};
use serde::{Deserialize, Serialize};
#[derive(Serialize, Deserialize, JsonSchema, ToAsyncApiMessage)]
#[serde(tag = "type")]
pub enum TimestampedMessage {
/// Event with timestamp
Event {
timestamp: DateTime<Utc>, // RFC3339 format
created_at: NaiveDateTime, // ISO8601 without timezone
message: String,
},
}Cargo.toml configuration:
[dependencies]
asyncapi-rust = "0.4"
chrono = { version = "0.4", features = ["serde"] }
schemars = { version = "1.1", features = ["derive", "chrono04"] }The chrono04 feature in schemars enables proper JSON schema generation for chrono datetime types. Without this feature, you would need to use #[schemars(skip)] and lose schema information for datetime fields.
Create a separate binary in your project to generate AsyncAPI specs:
// bin/generate-asyncapi.rs
use my_project::MyApi;
use asyncapi_rust::AsyncApi;
fn main() {
let spec = MyApi::asyncapi_spec();
let json = serde_json::to_string_pretty(&spec)
.expect("Failed to serialize spec");
std::fs::write("docs/asyncapi.json", json)
.expect("Failed to write spec file");
println!("✅ Generated docs/asyncapi.json");
}Run with:
cargo run --bin generate-asyncapiBenefits:
- Simple to implement and use
- Works with any build system
- Can commit generated spec to git for CI/CD
- Easy to integrate into workflows
You can include the generated spec in your crate's documentation:
#[doc = include_str!("../docs/asyncapi.json")]
#[derive(AsyncApi)]
#[asyncapi(title = "My API", version = "1.0.0")]
struct MyApi;This embeds the AsyncAPI specification directly in your rustdoc output, making it accessible alongside your Rust API documentation.
Workflow:
- Generate the spec file:
cargo run --bin generate-asyncapi - Build docs:
cargo doc - The AsyncAPI spec will be visible in the rustdoc for
MyApi
A cargo-asyncapi plugin for automatic spec generation is planned for a future release. This would allow:
cargo asyncapi generate
cargo asyncapi serve # Start AsyncAPI UI viewer- Core macro implementation
- actix-ws integration
- axum integration
- Binary message support
- Embedded AsyncAPI UI
- Additional framework support (tonic/gRPC, Rocket, Warp)
- Cargo plugin (
cargo-asyncapi) for automated spec generation - ~98% test coverage (measured by cargo-tarpaulin)
Contributions are welcome! Please see CONTRIBUTING.md for guidelines.
Licensed under either of:
- Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.
There has been a lot of discussion in the Rust community about usage of AI and LLMs. This project has been implemented with the assistance of Claude Code, but it is not vibe-coded. In a few years, using AI Tools will be common practice and these arguments will seem as quaint as those made decades ago against the use of IDEs. A human has designed this project and reviewed all code.
Inspired by:
- utoipa - OpenAPI code generation for Rust
- AsyncAPI Initiative - AsyncAPI specification
Author: Mark Lilback (mark@lilback.com) Repository: https://github.com/mlilback/asyncapi-rust