Crate gestura_core_llm 

Feature-gated LLM provider implementations and shared provider abstractions.

gestura-core-llm is the domain crate behind Gestura’s provider layer. It defines the common LlmProvider trait, shared response/token models, model listing helpers, default model catalogs, and the concrete provider implementations used by the runtime.

Supported providers

Provider implementations are enabled with Cargo features and currently cover:

• OpenAI
• Anthropic
• Grok (xAI)
• Gemini
• Ollama (local)

Design role

This crate owns provider-specific HTTP behavior and response normalization. Higher-level concerns such as configuration-driven provider selection, runtime overrides, and pipeline orchestration remain in gestura-core.

The stable public import path for most consumers remains gestura_core::llm_provider::*.

Shared abstractions

• LlmProvider: async provider interface used by the runtime
• LlmCallResponse: normalized response with text, usage, and tool calls
• TokenUsage: provider-agnostic token accounting and estimated cost data
• ToolCallInfo: normalized native function/tool call representation
• default_models, model_listing, token_tracker: support modules for model defaults, discovery, and token accounting

Native tool calling

Where providers support it, Gestura normalizes native function/tool calling into a common ToolCallInfo representation so the pipeline can process tool calls consistently across providers.

Feature-gated workspace design

This crate is intentionally feature-gated so applications can compile only the providers they need. That keeps optional integrations isolated and makes the workspace easier to reason about in cargo doc and CI.

Modules

default_models

Centralized default AI model constants for all providers.

model_capabilities

Dynamic model capabilities discovery and caching.

model_discovery

Dynamic model metadata discovery via provider APIs.

model_listing

Dynamic and static model listing for all LLM providers.

openai

OpenAI model capability and endpoint routing helpers.

token_tracker

Token usage tracking and cost estimation for Gestura

Structs

AgentContext

Context hints for provider selection (agent, tenant, etc.)

AnthropicProvider

HTTP-based Anthropic Claude provider

GeminiProvider

HTTP-based Google Gemini provider (Generative Language API).

GrokProvider

HTTP-based Grok (xAI) provider (OpenAI-compatible endpoint)

LlmCallResponse

Response from an LLM call including token usage

OllamaProvider

HTTP-based Ollama local provider

OpenAiProvider

HTTP-based OpenAI completion provider

TokenUsage

Token usage information from an LLM API call

ToolCallInfo

A structured tool call returned by the LLM when using native function calling.

UnconfiguredProvider

A provider that returns an error when no real provider is configured. Used when config is missing or invalid.

Traits

LlmProvider

Unified LLM interface (async)

Functions

unconfigured_provider

Create an unconfigured provider that returns an error when called. Used when a provider is not properly configured.