Skip to main content

Python SDK

The ARP Standard Python packages are generated from the v1 OpenAPI contracts.

They are intentionally split into three packages:

  • Models: arp-standard-model
  • Server base (FastAPI base servers): arp-standard-server
  • Clients (typed HTTP clients): arp-standard-client
Standard vs. implementation

These packages are generated from the ARP Standard spec and can be used with any conformant ARP Standard v1 service. They are not specific to JARVIS.

Packages​

  • Model PyPI distribution: arp-standard-model (imports: arp_standard_model)
  • Server base PyPI distribution: arp-standard-server (imports: arp_standard_server)
  • Client PyPI distribution: arp-standard-client (imports: arp_standard_client)
  • Python requirement: >=3.11
  • Package READMEs:
    • ARP_Standard/clients/python/README.md
    • ARP_Standard/models/python/README.md
    • ARP_Standard/kits/python/README.md (server base / kit)

Install​

Client + models (calling ARP services):

python3 -m pip install arp-standard-client

Server base + models (implementing ARP services):

python3 -m pip install arp-standard-server

Clients​

One facade client is generated per ARP Standard v1 service:

  • arp_standard_client.run_gateway.RunGatewayClient
  • arp_standard_client.run_coordinator.RunCoordinatorClient
  • arp_standard_client.atomic_executor.AtomicExecutorClient
  • arp_standard_client.composite_executor.CompositeExecutorClient
  • arp_standard_client.node_registry.NodeRegistryClient
  • arp_standard_client.selection.SelectionClient
  • arp_standard_client.pdp.PdpClient

Each client takes a base_url that points at the corresponding service endpoint.

Basic usage (Run Gateway example)​

This example checks health and starts a run:

from arp_standard_client.run_gateway import RunGatewayClient
from arp_standard_model import (
    NodeTypeRef,
    RunGatewayHealthRequest,
    RunGatewayStartRunRequest,
    RunStartRequest,
)

# If you started the pinned stack via `JARVIS_Release` Docker Compose, Run Gateway is exposed on `8081`.
gateway = RunGatewayClient(base_url="http://127.0.0.1:8081")

health = gateway.health(RunGatewayHealthRequest())
print(health.model_dump(exclude_none=True))

run = gateway.start_run(
    RunGatewayStartRunRequest(
        body=RunStartRequest(
            root_node_type_ref=NodeTypeRef(node_type_id="jarvis.composite.planner.general", version="0.3.3"),
            input={"goal": "Generate a UUID, then return it."},
        )
    )
)
print(run.model_dump(exclude_none=True))

Errors​

The facade clients raise arp_standard_client.errors.ArpApiError when the API returns an ErrorEnvelope.

Server base (FastAPI)​

If you’re implementing an ARP Standard service, use arp-standard-server so your server:

  • stays aligned with the OpenAPI contract (request/response envelopes),
  • returns ARP ErrorEnvelopes consistently (instead of framework-specific payloads),
  • exposes required endpoints (health/version) “conformant-by-construction”.

Each ARP service has a matching base class under arp_standard_server.<service>.

Generate locally​

If you want to see the code gen pipeline run locally, run these from the ARP_Standard repository root:

python3 -m pip install -r tools/codegen/python/model/requirements.txt
python3 -m pip install -r tools/codegen/python/client/requirements.txt
python3 -m pip install -r tools/codegen/python/server/requirements.txt
python3 tools/codegen/python/model/generate.py
python3 tools/codegen/python/client/generate.py
python3 tools/codegen/python/server/generate.py

See also: ARP_Standard/tools/codegen/python/README.md.