docs(sdk): explain OpenRouter model discovery

[enyst]

• I have read and reviewed the documentation changes to the best of my ability.
• If the change is significant, I have run the documentation site locally and confirmed it renders as expected.

Summary of changes

Document that the Agent Server keeps curated OpenRouter routes discoverable independently of LiteLLM's catalog. Explain the returned model prefix and existing runtime context-limit discovery.

Corresponding SDK fix: OpenHands/software-agent-sdk#5703

Validation: git diff --check and mint broken-links passed locally (Node 24). The PR's Mintlify internal-links CI also passed.

[all-hands-bot]

Docs review - content is accurate; PR is still draft

Read the one-file change (sdk/arch/llm.mdx) against the SDK fix in OpenHands/software-agent-sdk#5703. The paragraph is accurate: the Agent Server's /api/llm/models?provider=openrouter returns the curated routes even when LiteLLM's catalog omits them, identifiers keep the openrouter/ prefix, and context limits come from OpenRouter's endpoints API. No changes requested.

Two small things:

1. The PR is still marked draft, so it cannot merge as-is. Also, since the SDK fix's own CI is still red on its PR-description check, it is worth holding this until #5703 is ready so the doc does not land ahead of the behavior it describes.
2. "Returned identifiers keep the openrouter/ prefix, such as openrouter/deepseek/deepseek-chat, so they can be passed directly to LLM" is slightly imprecise - the returned string is the model id; you still configure the provider/key on LLM. Optional, but a word on that would preempt confusion.

This review was created by an AI agent (OpenHands) on behalf of the user.

[all-hands-bot]

Docs review - accurate; approving

Re-checked the one-file change (sdk/arch/llm.mdx) against the SDK fix. It is accurate and matches the now-merged behavior in OpenHands/software-agent-sdk#5703: /api/llm/models?provider=openrouter keeps the curated routes discoverable even when LiteLLM's catalog omits them, identifiers retain the openrouter/ prefix, and context limits come from OpenRouter's endpoints API. The example openrouter/deepseek/deepseek-chat is a real route in VERIFIED_OPENROUTER_MODELS, and LLM already knows the openrouter/ provider so the prefixed id is usable directly.

CI is green on the head: Check Internal Links (Mintlify) passed; Check External Links is skipped. Thanks for un-drafting it now that the SDK side has landed.

Tiny optional nit (not blocking): "so they can be passed directly to LLM" is slightly loose - the returned string is the model id; the provider/key still come from LLM config. A couple of words would preempt confusion, but the current wording is fine.

This review was created by an AI agent (OpenHands) on behalf of the user.