Dial a Service

There are three ways to call an Aster service, depending on your situation:

Approach	When to use	Needs local types?
Typed client	Application code, CI pipelines, production services	Yes
Aster shell	Exploration, debugging, ad-hoc calls, ops	No
Raw transport	Custom tooling, other-language FFI, testing	No

1. Typed client (recommended for application code)

The typed client gives you IDE autocomplete, type checking, and compile-time safety. You need the service's Python type definitions (the @service-decorated class and its request/response dataclasses).

import asyncio
from aster import AsterClient
from my_services import HelloService, HelloRequest

async def main():
    async with AsterClient(endpoint_addr="<base64 NodeAddr>") as c:
        hello = await c.client(HelloService)
        resp = await hello.say_hello(HelloRequest(name="World"))
        print(resp.message)  # Hello, World!

asyncio.run(main())

The consumer needs:

• The producer's endpoint address -- printed on startup, or set via ASTER_ENDPOINT_ADDR
• The service type definitions -- the same @service class (or a compatible package) as the producer

How it works

1. AsterClient connects to the producer's admission endpoint
2. The admission response lists available services (name, version, contract_id)
3. client.client(HelloService) matches the local class to the remote service by name + version
4. A QUIC connection opens on aster/1 and a typed ServiceClient is returned
5. Method calls serialize via Fory XLANG → send frame → receive frame → deserialize

Getting the types

In production, service authors publish a client package containing the type definitions:

pip install billing-service-types  # published by the service team

from billing_types import BillingService, InvoiceRequest

For internal services, the types often live in a shared module imported by both producer and consumer.

Identity and credentials

For dev mode (open gate), just set the address:

ASTER_ENDPOINT_ADDR=<addr> python consumer.py

For production, use an .aster-identity file:

# Generate with the CLI:
aster enroll node --profile prod --role consumer --name my-consumer

# Then just run -- the identity file provides everything:
ASTER_ENDPOINT_ADDR=<addr> python consumer.py

See AsterClient for the full Python API reference.

2. Aster shell (for exploration and debugging)

The Aster shell connects to any peer and lets you browse services, inspect contracts, and invoke methods interactively -- without needing the Python type definitions locally.

# Connect to a producer
aster shell <endpoint-addr-b64>

# Or explore in demo mode (no live peer needed)
aster shell --demo

What the shell can do

producer:/$ ls
  blobs/     services/     gossip/

producer:/$ cd services
producer:/services$ ls
  Service         Methods    Version    Pattern
  HelloService          1         v1    shared
  TaskRunner            3         v2    shared

producer:/services$ cd HelloService
producer:/services/HelloService$ ls
  Method       Pattern    Signature
  say_hello    unary      (HelloRequest) -> HelloResponse

producer:/services/HelloService$ ./say_hello name="World"
-> HelloService.say_hello(name='World')
(42ms)
{
  "message": "Hello, World!"
}

Dynamic invocation without local types

The shell does not just discover methods -- it can invoke them too, without any local type definitions. When connected to a peer, the shell reads contract manifests and synthesizes wire-compatible types at runtime. Field names, types, defaults, and Fory wire tags are all extracted from the manifest, so the shell can build correctly serialized requests on the fly. This means you can explore AND invoke any service from the shell without installing the service's client package.

How it discovers methods

The shell reads the service contract from the producer's registry:

1. Admission response includes a registry ticket (read-only iroh-docs share)
2. The shell joins the registry doc and reads ArtifactRef entries
3. Each ArtifactRef points to a blob collection containing manifest.json
4. The manifest has full method schemas: names, patterns, field definitions with types and defaults
5. Tab completion and interactive prompting work from this metadata

Shell features

• Filesystem-like navigation -- cd, ls, pwd, cat, save
• Tab completion -- service names, method names, argument names
• Interactive prompting -- call a method with no args and it prompts for each field
• All streaming patterns -- server-stream displays values as they arrive, bidi shows split input/output
• Contract introspection -- describe shows the full contract tree
• Rich output -- tables, syntax-highlighted JSON, progress bars
• Session subshell -- session Analytics opens a dedicated session for session-scoped services

CLI equivalents

Every shell command has a non-interactive equivalent for scripting:

aster service ls <peer>
aster service describe <peer> HelloService
aster blob ls <peer>
aster blob cat <peer> <hash>

See the CLI Reference for the full command reference.

3. Raw transport (for custom tooling)

For advanced use cases -- custom tooling, testing, or building clients in languages without a typed binding -- you can use the transport layer directly:

from aster import IrohTransport, ForyCodec, SerializationMode

# Open a QUIC connection
transport = IrohTransport(connection)

# Invoke by service and method name
result = await transport.unary(
    service="HelloService",
    method="say_hello",
    request=HelloRequest(name="World"),
)

The raw transport still requires serializable request/response types. For truly schemaless invocation, use the shell or fetch the manifest and build payloads from the field definitions.

Admission flow (all approaches)

Regardless of which approach you use, every consumer goes through the same admission flow:

The admission response gives the consumer:

• services -- list of available services with name, version, contract_id
• registry_ticket -- read-only doc ticket for full contract metadata
• root_pubkey -- for independent credential verification

Even in dev mode (open gate), the admission handshake runs to discover services.

Error handling

All three approaches produce the same error types:

Error	Meaning
PermissionError	Admission denied (invalid credential, wrong root key)
LookupError	Service not offered by this producer
RpcError	Call failed (check e.code: NOT_FOUND, DEADLINE_EXCEEDED, INTERNAL, etc.)
ConnectionError	QUIC connection lost

from aster import RpcError, StatusCode

try:
    resp = await hello.say_hello(HelloRequest(name=""))
except RpcError as e:
    if e.code == StatusCode.DEADLINE_EXCEEDED:
        print("Call timed out")
    elif e.code == StatusCode.PERMISSION_DENIED:
        print("Not authorized for this method")