RFC: Publish dynamo-protocols and dynamo-parsers as standalone crates

[ishandhanani]

Summary

Publish dynamo-protocols and dynamo-parsers as standalone, reusable crates on crates.io so that the broader Rust inference ecosystem can leverage battle-tested protocol types and streaming parsers without pulling in all of Dynamo.

Architecture

dynamo-protocols depends on upstream async-openai v0.34 for base OpenAI types and layers inference-serving extensions on top.

dynamo-protocols
├── async-openai v0.34 (upstream, re-exported)
│   ├── chat types: CompletionUsage, FunctionCall, Role, Prompt, ...
│   ├── completion types: CreateCompletionResponse
│   ├── embedding types: full re-export
│   ├── responses types: full re-export
│   └── image types: full re-export
├── inference-serving extensions (locally defined)
│   ├── ChatCompletionResponseMessage (multimodal content + reasoning_content)
│   ├── ChatChoiceStream (stop_reason from backend)
│   ├── CreateChatCompletionRequest (mm_processor_kwargs, extended stream_options)
│   ├── ReasoningContent, StopReason, ChatCompletionMessageContent
│   ├── VideoUrl, AudioUrl (multimodal input)
│   └── ImageUrl (url::Url + uuid tracking)
└── anthropic/ (full Anthropic Messages API)
    ├── AnthropicCreateMessageRequest, AnthropicMessageResponse
    ├── AnthropicStreamEvent, CacheControl, SystemContent
    └── Token counting, error types

PR Stack

PR 1: Remove nvext from base types + extract Anthropic types

#7564 | idhanani/remove-nvext-from-base-types | Open

PR 2: Rename to dynamo-protocols

#7565 | idhanani/rename-to-dynamo-protocols | Open

PR 3: Switch to upstream async-openai + remove fork

#7625 | idhanani/protocols-cleanup | Open

• Add async-openai = "0.34" (types-only features)
• Re-export upstream types for chat, completion, embedding, responses, images
• Keep ~15 locally-defined types for inference-serving extensions
• Remove utoipa/ToSchema
• Delete entire forked HTTP client, API handlers, and unused type modules
• -16,811 lines total across the PR stack

Remaining work

• cargo package validation and publish to crates.io

Types retained locally

Type	Extension	Rationale
ChatCompletionResponseMessage	content: ChatCompletionMessageContent, reasoning_content	Multimodal response content + reasoning output
ChatCompletionStreamResponseDelta	Same as above	Streaming variant
ChatChoice / ChatChoiceStream	stop_reason: Option<StopReason>	Backend-reported stop condition
CreateChatCompletionRequest	mm_processor_kwargs	vLLM multimodal processor config
ChatCompletionStreamOptions	continuous_usage_stats	Per-chunk usage reporting
ChatCompletionRequestAssistantMessage	reasoning_content: Option<ReasoningContent>	KV cache-correct reasoning
ChatCompletionRequestUserMessageContentPart	VideoUrl, AudioUrl variants	Multimodal input
ImageUrl	url::Url, uuid field	Strongly-typed URLs, asset tracking
CreateCompletionRequest	prompt_embeds, echo validation	Pre-computed embeddings
All anthropic/ types	Fully custom	Anthropic Messages API

Usage

[dependencies]
dynamo-protocols = "1.0"
dynamo-parsers = "1.0"

use dynamo_protocols::types::{
    CreateChatCompletionRequest,
    CreateChatCompletionStreamResponse,
    CompletionUsage,
    ReasoningContent,
};
use dynamo_protocols::types::responses::{CreateResponse, Response};
use dynamo_protocols::types::anthropic::{
    AnthropicCreateMessageRequest,
    AnthropicMessageResponse,
    CacheControl,
};