feat: add OrcaRouter as LLM provider for workflow generation

[zsanig22-dotcom]

Summary

Workflow Use's generate-workflow command already accepts --agent-model/--extraction-model options, but those models were never wired up — the agent and extraction LLMs were always built as ChatBrowserUse(model='bu-latest'). This PR makes those options effective and adds --llm-provider, with orcarouter as a first-class provider backed by OrcaRouter.

OrcaRouter is an OpenAI-compatible AI gateway built for both models and agents. Like OpenRouter, it exposes a provider/model namespace across many models — but it also combines adaptive routing, automatic failover, zero-markup inference, observability, guardrails, and agent-tool governance behind the same endpoint. Adding orcarouter as a first-class provider means this project's users can use that stack directly, without treating OrcaRouter as an anonymous custom base URL. 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.

The implementation mirrors how browser_use.llm.openai.ChatOpenAI is already used by the ecosystem: an OpenAI-compatible client pointed at OrcaRouter's https://api.orcarouter.ai/v1 endpoint with ORCAROUTER_API_KEY. The default (browser-use) path is unchanged.

Verification

• ruff check and ruff format --check pass on workflows/ (matches CI).
• Live-tested the new orcarouter code path against https://api.orcarouter.ai/v1 with model=orcarouter/auto → HTTP 200, completion returned (routed to gemini-3.1-flash-lite).

Discord: discord.gg/YEubt8enRA · X: https://x.com/OrcaRouter

I'm an engineer on the OrcaRouter team.

---

Summary by cubic

Makes the --agent-model and --extraction-model options in generate-workflow actually take effect, and adds a new --llm-provider option with orcarouter as a first-class provider backed by OrcaRouter's OpenAI-compatible gateway.

Previously these model options were ignored; the agent and extraction LLMs were always built with ChatBrowserUse(model='bu-latest'). The default browser-use path is unchanged.

• The orcarouter provider routes workflow generation through OrcaRouter's gateway for adaptive routing and failover.
• Users need to set ORCAROUTER_API_KEY in the environment to use the orcarouter provider.
• --agent-model and --extraction-model now apply to the selected provider.

^{Written for commit 7bd43ba. Summary will update on new commits.}

[cubic-dev-ai]

3 issues found across 3 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="workflows/cli.py">

<violation number="1" location="workflows/cli.py:2311">
P2: When `--llm-provider` contains a typo or unsupported value, this fallback silently uses Browser-Use instead of rejecting the provider. Validate `browser-use` explicitly and raise a CLI parameter error for every other value.</violation>

<violation number="2" location="workflows/cli.py:2315">
P2: When `ORCAROUTER_API_KEY` is unset, `os.getenv` returns None, which ChatOpenAI drops from its client params and AsyncOpenAI then fills by falling back to `OPENAI_API_KEY`. The result is your OpenAI key being sent to OrcaRouter's endpoint, or a confusing client error if no key is set. Validate the key is non-empty before building the orcarouter LLM.</violation>

<violation number="3" location="workflows/cli.py:2317">
P2: When the default `browser-use` provider is selected, `--agent-model` and `--extraction-model` are ignored because this branch always creates `bu-latest`. Pass `model` to `ChatBrowserUse` so the documented custom-model options work for the default provider.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

P2: When the default browser-use provider is selected, --agent-model and --extraction-model are ignored because this branch always creates bu-latest. Pass model to ChatBrowserUse so the documented custom-model options work for the default provider.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At workflows/cli.py, line 2317:

<comment>When the default `browser-use` provider is selected, `--agent-model` and `--extraction-model` are ignored because this branch always creates `bu-latest`. Pass `model` to `ChatBrowserUse` so the documented custom-model options work for the default provider.</comment>

<file context>
@@ -2300,6 +2300,23 @@ def generate_csv_template_command(
+			base_url='https://api.orcarouter.ai/v1',
+			api_key=os.getenv('ORCAROUTER_API_KEY'),
+		)
+	return ChatBrowserUse(model='bu-latest')
+
+
</file context>

Suggested change

	return ChatBrowserUse(model='bu-latest')
	return ChatBrowserUse(model=model)

[cubic-dev-ai]

P2: When --llm-provider contains a typo or unsupported value, this fallback silently uses Browser-Use instead of rejecting the provider. Validate browser-use explicitly and raise a CLI parameter error for every other value.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At workflows/cli.py, line 2311:

<comment>When `--llm-provider` contains a typo or unsupported value, this fallback silently uses Browser-Use instead of rejecting the provider. Validate `browser-use` explicitly and raise a CLI parameter error for every other value.</comment>

<file context>
@@ -2300,6 +2300,23 @@ def generate_csv_template_command(
+	the existing browser-use cloud behavior (the ``bu-latest`` alias is resolved
+	by the gateway to the current premium model).
+	"""
+	if llm_provider == 'orcarouter':
+		return ChatOpenAI(
+			model=model,
</file context>

Suggested change

	if llm_provider == 'orcarouter':
	return ChatOpenAI(
	model=model,
	base_url='https://api.orcarouter.ai/v1',
	api_key=os.getenv('ORCAROUTER_API_KEY'),
	)
	return ChatBrowserUse(model='bu-latest')
	if llm_provider == 'orcarouter':
	return ChatOpenAI(
	model=model,
	base_url='https://api.orcarouter.ai/v1',
	api_key=os.getenv('ORCAROUTER_API_KEY'),
	)
	elif llm_provider == 'browser-use':
	return ChatBrowserUse(model='bu-latest')
	raise typer.BadParameter(f'Unsupported LLM provider: {llm_provider}')

[cubic-dev-ai]

P2: When ORCAROUTER_API_KEY is unset, os.getenv returns None, which ChatOpenAI drops from its client params and AsyncOpenAI then fills by falling back to OPENAI_API_KEY. The result is your OpenAI key being sent to OrcaRouter's endpoint, or a confusing client error if no key is set. Validate the key is non-empty before building the orcarouter LLM.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At workflows/cli.py, line 2315:

<comment>When `ORCAROUTER_API_KEY` is unset, `os.getenv` returns None, which ChatOpenAI drops from its client params and AsyncOpenAI then fills by falling back to `OPENAI_API_KEY`. The result is your OpenAI key being sent to OrcaRouter's endpoint, or a confusing client error if no key is set. Validate the key is non-empty before building the orcarouter LLM.</comment>

<file context>
@@ -2300,6 +2300,23 @@ def generate_csv_template_command(
+		return ChatOpenAI(
+			model=model,
+			base_url='https://api.orcarouter.ai/v1',
+			api_key=os.getenv('ORCAROUTER_API_KEY'),
+		)
+	return ChatBrowserUse(model='bu-latest')
</file context>