Struct ProviderCapabilities 

pub struct ProviderCapabilities {
    pub streaming_tool_calls: bool,
    pub named_sse_events: bool,
    pub reasoning_content: bool,
    pub image_input: bool,
    pub system_as_top_level: bool,
    pub ndjson_streaming: bool,
    pub tool_support: bool,
    pub structured_output: bool,
    pub requires_version_header: bool,
    pub supports_parallel_tool_calls: bool,
    pub max_context_tokens: Option<u32>,
    pub tokenizer_name: Option<&'static str>,
}

Protocol-level features that differ across LLM providers.

Fields

streaming_tool_calls: bool

Supports incremental streaming tool call deltas (OpenAI delta.tool_calls). When false, tool calls are only available in final message (Anthropic block events).

named_sse_events: bool

Uses named SSE event types (Anthropic event: content_block_start etc.) vs bare data: lines (OpenAI-compatible).

reasoning_content: bool

Supports reasoning_content / thinking output in streaming responses (Qwen3, DeepSeek, OpenAI o-series).

image_input: bool

Supports image inputs (multimodal content parts).

system_as_top_level: bool

System prompt sent as a top-level request field (Anthropic system) vs as a message in the messages array with role: "system".

ndjson_streaming: bool

Uses NDJSON (one JSON object per line, \n delimited) for streaming instead of standard SSE data: lines (Ollama).

tool_support: bool

Supports function/tool definitions in requests.

structured_output: bool

Supports structured output (response_format with JSON Schema).

requires_version_header: bool

Requires a provider-specific version header (Anthropic anthropic-version).

supports_parallel_tool_calls: bool

Supports parallel tool calls (multiple tools requested in a single response). OpenAI, Anthropic, and most cloud providers support this; Ollama and local models typically do not.

max_context_tokens: Option<u32>

Known maximum context window in tokens for this provider’s default model. None means unknown.

tokenizer_name: Option<&'static str>

Tokenizer name or identifier for accurate token counting (e.g. "cl100k_base" for GPT-4, "o200k_base" for GPT-4o). None means the tokenizer is unknown; callers should fall back to heuristic counting.

Implementations

impl ProviderCapabilities

pub const fn openai_compatible() -> ProviderCapabilities

Default capabilities for OpenAI and OpenAI-compatible providers (DashScope, DeepSeek, Moonshot, Zhipu, etc.).

pub const fn anthropic() -> ProviderCapabilities

Capabilities for Anthropic Messages API.

pub const fn ollama() -> ProviderCapabilities

Capabilities for local Ollama.

pub fn from_provider_name(name: &str) -> ProviderCapabilities

Resolve capabilities from a provider name string.

Trait Implementations

impl Clone for ProviderCapabilities

fn clone(&self) -> ProviderCapabilities

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ProviderCapabilities

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl PartialEq for ProviderCapabilities

fn eq(&self, other: &ProviderCapabilities) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Copy for ProviderCapabilities

impl Eq for ProviderCapabilities

impl StructuralPartialEq for ProviderCapabilities