feat(llms): add OrcaRouter named provider

[JinhaoSong322]

Summary

Add OrcaRouter as a named LLM provider in Lagent, mirroring the existing GPTAPI / AsyncGPTAPI OpenAI-compatible wrappers.

OrcaRouter is a gateway that fronts open-weight models from many vendors through a single OpenAI-compatible endpoint. It also runs gateway-level, zero-trust security for AI agents on the same endpoint — screening every prompt/response and governing every tool call on a default-deny basis, with no application code changes.

What changed

• lagent/llms/orcarouter.py (new): OrcaRouterAPI and AsyncOrcaRouterAPI, subclasses of GPTAPI / AsyncGPTAPI with OrcaRouter defaults:
  • base url https://api.orcarouter.ai/v1/chat/completions
  • default model orcarouter/auto (auto-routed open-weight models on the gateway)
  • reads ORCAROUTER_API_KEY (keys start with sk-orca-) instead of OPENAI_API_KEY
  • accepts gateway aliases (orcarouter/*) and vendor-qualified model names (vendor/model, e.g. deepseek/deepseek-v4-pro) via the generic Chat Completions payload
• lagent/llms/__init__.py: export the new wrappers.
• docs/en/get_started/quickstart.md: document the "Using OrcaRouter" section.
• tests/test_llms/test_orcarouter.py (new): 11 unit tests covering defaults, env-key resolution, payload generation (gateway alias / vendor-qualified / json mode / unsupported model), and a mocked chat call.

Verification

• flake8 clean on changed files; py_compile passes.
• Unit tests: 11/11 passed.
• Full suite: identical results to clean main (7 pre-existing environment/network-dependent failures in tests/test_actions/*; tests/test_agents/test_rewoo.py fails collection on clean main too due to a refactored-out llm_qa module — unrelated to this change).
• L3 live test against https://api.orcarouter.ai/v1 with a real key, through the new provider path:
  • sync chat (orcarouter/auto) → ORCA-OK
  • async chat → ORCA-OK
  • vendor-qualified model deepseek/deepseek-v4-pro → VENDOR-OK
  • json mode payload generation verified

Example

import os
os.environ['ORCAROUTER_API_KEY'] = 'sk-orca-...'

from lagent.llms import OrcaRouterAPI
llm = OrcaRouterAPI(model_type='orcarouter/auto', retry=5, max_new_tokens=2048)
print(llm.chat([{'role': 'user', 'content': 'Hello!'}]))

Disclosure: I'm an engineer on the OrcaRouter team.