feat(models): add Portkey AI gateway model provider

[kyesw]

Summary

• Add PortkeyModel integration that routes requests through the Portkey AI gateway, enabling observability, caching, fallbacks, and load balancing across any LLM provider
• The gateway automatically infers the provider from model_id, so only api_key and model_id are required:

  model = PortkeyModel(api_key="pk-...", model_id="gpt-4o")

• Extends OpenAIModel since Portkey normalizes all provider responses into the OpenAI-compatible format
• Add portkey-ai as optional dependency (pip install strands-agents[portkey])

Files changed

File	Change
src/strands/models/portkey.py	New PortkeyModel implementation
src/strands/models/__init__.py	Lazy loading registration
tests/strands/models/test_portkey.py	20 unit tests
pyproject.toml	Optional dependency + all extra

Usage examples

from strands import Agent
from strands.models.portkey import PortkeyModel

# OpenAI
model = PortkeyModel(api_key="pk-...", model_id="gpt-4o")

# Anthropic — just change the model_id
model = PortkeyModel(api_key="pk-...", model_id="claude-sonnet-4-20250514")

# Google
model = PortkeyModel(api_key="pk-...", model_id="gemini-2.0-flash")

# With Portkey config for routing/fallbacks/load balancing
model = PortkeyModel(api_key="pk-...", config="cf-xxx", model_id="gpt-4o")

agent = Agent(model=model)
agent("Hello!")

Test plan

• hatch fmt --formatter — passed
• hatch fmt --linter — passed
• hatch test -- tests/strands/models/test_portkey.py — 20/20 passed
• Manual verification with Portkey gateway using Bedrock model