refactor(native): separate request data from execution context

[yujonglee-berri]

TLDR

Problem this solves:

• Requests mixed payloads with incomplete execution context

How it solves it:

• Separate typed request, options, and context

User Flow

Before: the SDK request can violate this layer contract.

1. They send a traced SDK request with route-specific options.
2. Some identity or capability facts are unavailable downstream.

After: the same request follows one predictable contract.

1. They send the same traced SDK request.
2. Call ID, trace ID, model, attribution, mode, and capabilities survive.

Relevant issues

• Stack root: refactor(rust): remove per-request enablement arguments #39928

Linear ticket

Pre-Submission checklist

• I have added meaningful tests
• Focused tests covering this change pass locally
• All required CI/CD checks pass
• This PR is isolated to one stack responsibility
• Greptile Confidence Score is at least 4/5

Stack

39928 -> 39913 -> 39923 -> 39941 -> 39939 -> 39484 -> 39986 -> 40054 -> 40055 -> 40056 -> 39765

Layer 4 of 11.

Screenshots / Proof of Fix

After (c46fd72df65801f31265abc978dab5fc6142c33f)

1. Focused request-context and marshaling tests: passed
2. UV_PYTHON=3.12 make check: passed
3. Final stack Rust workspace and strict Clippy: passed

No paid-provider calls were used for deterministic validation.

Type

Refactoring

Caveats (if any)

Medium

• Opaque input consumption lands in later stack layers.

Final Attestation

• Tests cover this layer intended contracts and edge cases

[github-actions]

LiteLLM Rust workflow

Workflow completed successfully for 905c5cb7bb4cd818114defa987a8864ea789603b

View workflow run

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_native_request_context (12c02dd) with litellm_native_dispatch_error_fixes (7ed2207)}

[codecov]

Codecov Report

❌ Patch coverage is 94.15205% with 10 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...tellm/llms/bedrock/audio_transcription/__init__.py	61.53%	5 Missing ⚠️
litellm/rust_bridge/ocr.py	58.33%	5 Missing ⚠️

📢 Thoughts on this report? Let us know!

[greptile-apps]

Greptile Summary

This PR separates endpoint payloads, execution options, request identity, attribution, and capability facts across the Python-to-Rust native boundary.

• Introduces typed Python and Rust request context/options structures.
• Migrates chat completions, messages, OCR, transcription, and Responses WebSocket adapters to the new boundary.
• Moves Bedrock and Vertex execution settings into provider-specific typed options.
• Updates bridge marshaling and context-preservation tests.
• Leaves the server-enabled Rust messages route using the old combined request shape, causing a compilation failure.

Confidence Score: 4/5

The PR is not safe to merge until the server-enabled Rust gateway messages route is updated to the new three-argument API and compiles.

One build-blocking API migration error remains: the gateway constructs a nonexistent MessagesRequest.options field and fails to pass RequestOptions separately to either messages entrypoint.

Files Needing Attention: litellm-rust/crates/ai-gateway/src/routes/messages/service.rs, litellm-rust/README.md

Important Files Changed

Filename	Overview
litellm-rust/crates/ai-gateway/src/routes/messages/service.rs	Migrates the gateway messages call incorrectly; the server feature cannot compile because options are embedded in MessagesRequest and omitted from entrypoint arguments.
litellm-rust/crates/core/src/messages/mod.rs	Changes messages entrypoints to receive request options and execution context as separate arguments.
litellm-rust/crates/core/src/request_context.rs	Adds typed call identity, attribution, and execution-capability context.
litellm-rust/crates/core/src/request_options.rs	Adds shared execution options with typed Bedrock, Anthropic, and Vertex configuration.
litellm-rust/crates/python-bridge/src/marshal.rs	Marshals the new Python request options and context records into Rust types.
litellm/rust_bridge/request.py	Defines immutable Python request, options, context, and provider-option records used by native calls.
litellm-rust/README.md	Documents the new native boundary but shows the same invalid MessagesRequest.options shape as the broken gateway call site.

_{Reviews (1): Last reviewed commit: "feat(native): preserve request identity ..." | Re-trigger Greptile}

[greptile-apps]

Messages route fails compilation

The server-enabled gateway constructs MessagesRequest with an options field, but that type only contains model and body. The messages and messages_stream functions also require RequestOptions as a separate argument. As a result, building litellm-ai-gateway with the server feature fails with E0560 and E0061. The README example uses the same invalid request shape and should be corrected as well.

Suggested change

	let request = MessagesRequest {
	model: provider_model,
	body,
	api_key: deployment.litellm_params.api_key.as_deref(),
	api_base: deployment.litellm_params.api_base.as_deref(),
	custom_llm_provider,
	extra_headers,
	timeout: None,
	options: RequestOptions {
	api_key: (deployment.litellm_params.api_key.as_deref()).map(|value| value.to_string()),
	api_base: (deployment.litellm_params.api_base.as_deref())
	.map(|value| value.to_string()),
	custom_llm_provider: (custom_llm_provider).map(|value| value.to_string()),
	extra_headers,
	timeout: None,
	..Default::default()
	},
	};
	if request.body.get("stream").and_then(Value::as_bool) == Some(true) {
	return messages_stream(request).await.map(MessagesResponse::Stream);
	return messages_stream(
	request,
	&LiteLlmRequestContext {
	..Default::default()
	},
	)
	.await
	.map(MessagesResponse::Stream);
	}
	let response = messages(request).await?;
	let response = messages(
	request,
	&LiteLlmRequestContext {
	..Default::default()
	},
	)
	.await?;
	let options = RequestOptions {
	api_key: (deployment.litellm_params.api_key.as_deref()).map(|value| value.to_string()),
	api_base: (deployment.litellm_params.api_base.as_deref()).map(|value| value.to_string()),
	custom_llm_provider: (custom_llm_provider).map(|value| value.to_string()),
	extra_headers,
	timeout: None,
	..Default::default()
	};
	let request = MessagesRequest {
	model: provider_model,
	body,
	};
	if request.body.get("stream").and_then(Value::as_bool) == Some(true) {
	return messages_stream(
	request,
	&options,
	&LiteLlmRequestContext {
	..Default::default()
	},
	)
	.await
	.map(MessagesResponse::Stream);
	}
	let response = messages(
	request,
	&options,
	&LiteLlmRequestContext {
	..Default::default()
	},
	)
	.await?;

[yujonglee-berri]

Closing; stack is being cleaned up.