fix(llm): omit parallel_tool_calls on tool-less requests

[devin-ai-integration]

Summary

Azure OpenAI rejects any chat.completions request that carries parallel_tool_calls without tools:

litellm.BadRequestError: AzureException BadRequestError - Invalid value for 'parallel_tool_calls': 'parallel_tool_calls' is only allowed when 'tools' are specified.

make_model_settings hardcodes parallel_tool_calls=False, and the openai-agents LiteLLM model forwards False even when tools=[], so every tool-less request fails on Azure. Since the scan preflight was added, that failure is fatal ("Scan failed") at startup. LiteLLM's drop_params does not drop this param (BerriAI/litellm#8980), so it must be omitted at the source.

 def make_model_settings(..., extra_headers=None,
+    has_tools: bool = True,
 ) -> ModelSettings:
     model_settings = ModelSettings(
-        parallel_tool_calls=False,
+        parallel_tool_calls=False if has_tools else None,  # None => omitted on the wire

The four tool-less call sites now pass has_tools=False: scan preflight (preflight_model_connection), dedupe-model warm-up, compaction _summarize, and finding dedupe _dedupe_model_settings. The agent loop is unchanged and keeps parallel_tool_calls=False.

Verified against a mock Azure endpoint enforcing the real invariant: before, the preflight sent {"parallel_tool_calls": false} with no tools and got the 400 above; after, the param is omitted and the preflight succeeds.

Link to Devin session: https://app.devin.ai/sessions/b7e52c0b8fbb4f148d2fb08fb00a1534
Requested by: @0xallam

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Greptile Summary

The PR conditionally omits parallel_tool_calls for requests without tools, preventing Azure OpenAI from rejecting startup preflight, warm-up, compaction, and finding-deduplication requests.

• Adds a has_tools argument to the shared model-settings builder while preserving existing behavior for tool-enabled agents.
• Marks all four known tool-less model call paths with has_tools=False.
• Adds unit coverage for both tool-enabled and tool-less settings.

Confidence Score: 5/5

The PR appears safe to merge, with every identified tool-less caller updated consistently and no concrete regression found.

The shared default remains unchanged for the tool-enabled agent loop, while all four reachable calls that pass an empty tools list now request omission of the Azure-incompatible parameter.

Important Files Changed

Filename	Overview
strix/core/inputs.py	Adds the centralized tool-presence switch and maps tool-less requests to parallel_tool_calls=None.
strix/interface/main.py	Marks the dedupe-model warm-up as tool-less, consistent with its empty tools list.
strix/interface/scan_setup.py	Marks the startup model preflight as tool-less, preventing Azure from rejecting the request.
strix/llm/compaction.py	Marks pure-text context summarization as tool-less without changing its token-limit settings.
strix/report/dedupe.py	Marks finding-deduplication model requests as tool-less, consistent with the empty tools list.
tests/test_inputs.py	Covers the settings-level behavior for both default tool-enabled calls and explicitly tool-less calls.

_{Reviews (1): Last reviewed commit: "fix(llm): omit parallel_tool_calls on to..." | Re-trigger Greptile}