Python SDK - Workload Identity Federation

For deployments that prefer not to manage static API keys, the Python SDK can authenticate via Workload Identity Federation (WIF). The SDK fetches a workload OIDC token from the running environment, exchanges it for a Nexus session JWT, and uses Authorization: Bearer <jwt> on every subsequent API call. WIF is opt-in. When NexusConfig.wif is None (the default), the SDK falls back to the static api_key flow and behaves exactly as in v0.1.0. Both NexusClient (sync) and AsyncNexusClient (asyncio) support WIF.

Enabling

from westyx_nexus import NexusClient, NexusConfig, WIFConfig

client = NexusClient.create(NexusConfig(
    base_url="https://blue-ocean-a5rx7.westyx.dev",
    wif=WIFConfig(enabled=True),  # provider="auto" by default
))

api_key becomes optional when WIF is enabled.

Security note (v0.9.0): base_url must use https:// - NexusClient.create / AsyncNexusClient.create reject plain-http endpoints (loopback/localhost excepted for development), since credentials travel on every request.

Supported providers

When provider is "auto" the SDK probes the environment in this order and picks the first provider whose credential material is actually present (v0.9.0 - file stats + live metadata probes, ~1 s timeout; earlier versions keyed on environment variables absent on real cloud nodes):

Provider constant	Auto-detect signal	Credential
`WIF_PROVIDER_KUBERNETES`	`/var/run/secrets/kubernetes.io/serviceaccount/token` exists	reads the projected service-account token file (rotated by kubelet)
`WIF_PROVIDER_AWS`	`AWS_WEB_IDENTITY_TOKEN_FILE` set and the file exists (EKS IRSA)	reads the IRSA token file at that path
`WIF_PROVIDER_AZURE` (Workload Identity)	`AZURE_FEDERATED_TOKEN_FILE` set and the file exists (AKS)	reads the projected federated token file
`WIF_PROVIDER_GCP`	`metadata.google.internal` reachable (live probe)	GETs `metadata.google.internal/computeMetadata/v1/.../identity?audience=<aud>` with `Metadata-Flavor: Google`
`WIF_PROVIDER_AZURE` (IMDS)	`169.254.169.254/metadata` reachable (live probe)	GETs an Azure IMDS managed-identity token - requires `audience="api://<client-id>"` (your app registration’s Application ID URI; the generic default is refused because Azure AD rejects it)
`WIF_PROVIDER_AWS_IAM`	not auto-detected - select explicitly (resolvable AWS credentials alone, e.g. a dev laptop’s `~/.aws/credentials`, are too weak a signal)	SigV4-signs an STS `GetCallerIdentity` request (never sent to AWS) that Nexus replays to prove your IAM role - works on ECS/Fargate, Lambda, and plain EC2

The OIDC providers are httpx-only; aws_iam uses botocore (the optional [aws-iam] extra) for the credential/region chain and the SigV4 signer. You can also pin a specific provider:

from westyx_nexus import WIF_PROVIDER_KUBERNETES

WIFConfig(enabled=True, provider=WIF_PROVIDER_KUBERNETES)

AWS IAM (non-EKS AWS compute)

ECS/Fargate tasks, Lambda functions, and plain EC2 instances have IAM credentials but no OIDC token, so the OIDC providers cannot serve them. The aws_iam provider closes this gap. It requires the optional botocore dependency:

pip install 'westyx-nexus-sdk[aws-iam]'

from westyx_nexus import WIF_PROVIDER_AWS_IAM

WIFConfig(
    enabled=True,
    provider=WIF_PROVIDER_AWS_IAM,
    aws_region="eu-central-1",  # optional - defaults to AWS_REGION / shared config / EC2 IMDS
)

The SDK SigV4-signs an STS GetCallerIdentity request with the standard AWS credential chain (task role, instance profile, env) - never sending it to AWS - and posts the signed headers + body to /v1/auth/token-exchange as {"provider":"aws_iam","aws_sts_request":...}. Nexus replays it against a pinned STS endpoint and matches the caller’s IAM role against an aws_iam trust policy (subject = arn:aws:iam::<account>:role/<name>). Nothing else to configure: the SDK signs your service’s own host (the base_url host) into X-Nexus-Server-ID inside the signature, and Nexus verifies it against the host the request arrived on - so a captured signed request is valid for that ONE service only. Selecting aws_iam without botocore installed raises an actionable error naming the extra.

Custom token source

Provide a token_source callable to bypass auto-detection. The callable may be sync (returns str) or async (returns Awaitable[str]); the async client awaits it automatically.

def my_token_source() -> str:
    return read_my_oidc_token()

WIFConfig(enabled=True, token_source=my_token_source)

async def my_async_source() -> str:
    return await fetch_oidc_token_async()

WIFConfig(enabled=True, token_source=my_async_source)

When token_source is set it takes precedence over provider.

Audience

GCP and Azure require an audience claim when issuing the OIDC token:

WIFConfig(
    enabled=True,
    provider="gcp",
    audience="westyx-nexus",
)

The default is "westyx-nexus" if unset. Azure IMDS is the exception: the generic default is refused with a clear error, because Azure AD rejects it as a resource. On the IMDS path (plain VM / App Service - no federated token file) you MUST set audience to your app registration’s Application ID URI (api://<client-id>). On AKS with Azure Workload Identity the projected federated token file ($AZURE_FEDERATED_TOKEN_FILE) is preferred automatically and no audience is needed.

Token exchange flow

OIDC token fetch - the configured token source returns a workload-identity JWT.

Token exchange - the SDK POSTs {"oidc_token": "<jwt>"} to /v1/auth/token-exchange. The backend validates the OIDC issuer/signature and returns:

{
  "session_token":  "<jwt>",
  "token_type":     "Bearer",
  "expires_in":     3600,
  "service_kind":   "backend",
  "wif_provider":   "kubernetes"
}

Session storage - the SDK caches the session token in memory and uses Authorization: Bearer <session> on all subsequent /sync and /stream requests.
Auto-refresh - ~60 s before expiry the SDK silently re-runs steps 1-2 from a background thread (sync) or task (async). The active SSE stream is not interrupted.
auth_expiring event - when the server emits this control event over SSE, the SDK pre-emptively refreshes the session before the server force-closes the stream.

Split httpx clients

The SDK uses two separate httpx.Client instances - one for short requests (/sync, /token-exchange, with timeout=cfg.timeout_seconds) and one for the long-lived SSE stream (with timeout=None). This prevents a consumer-supplied short timeout from killing the stream after a few seconds.

Project-level vs service-level trust policies

The Nexus backend supports trust policies at two scopes:

Scope	What it covers	Recommended for
Project-level	All services in the project share one policy	Most teams - no per-service cloud identity needed
Service-level	A specific policy bound to one service only	High-security services (payments, PII)

A project-level trust policy is less restrictive than a service-level one. Any workload that satisfies the policy can connect to any service in the project by targeting its base_url. For services handling sensitive data (payments, credentials, PII) consider a dedicated service-level trust policy.

Overview

Backend SDKs

Browser SDKs

OpenFeature Providers

Python SDK - Workload Identity Federation

Enabling

Supported providers

AWS IAM (non-EKS AWS compute)

Custom token source

Audience

Token exchange flow

Split httpx clients

Project-level vs service-level trust policies

​Enabling

​Supported providers

​AWS IAM (non-EKS AWS compute)

​Custom token source

​Audience

​Token exchange flow

​Split httpx clients

​Project-level vs service-level trust policies

Enabling

Supported providers

AWS IAM (non-EKS AWS compute)

Custom token source

Audience

Token exchange flow

Split httpx clients

Project-level vs service-level trust policies