embacle 0.28.2

LLM runner library — 12 AI CLI tools, ACP, and HTTP APIs (Gemini, Cohere, Groq, OpenRouter, OpenAI-compatible) as pluggable LLM providers with fallback chains, agent loop, guardrails, and cost tracking
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56

// ABOUTME: HTTP API providers — Gemini, Cohere, Groq, OpenRouter, OpenAI-compatible — and what they share
// ABOUTME: One HTTP error mapping, one SSE parser, one retry policy; each provider keeps its own vendor quirks
//
// SPDX-License-Identifier: Apache-2.0
// Copyright (c) 2026 dravr.ai

//! # HTTP API providers (feature `http-api`)
//!
//! Providers that speak to a vendor's HTTP API directly, each implementing
//! [`LlmProvider`](crate::types::LlmProvider):
//!
//! | Provider | Type | `name()` | Env (`from_env`) |
//! |---|---|---|---|
//! | Google Gemini | [`GeminiProvider`](gemini::GeminiProvider) | `gemini` | `GEMINI_API_KEY`, `GEMINI_DEFAULT_MODEL`, `GEMINI_MAX_RETRIES`, `GEMINI_INITIAL_RETRY_DELAY_MS`, `GEMINI_MAX_RETRY_DELAY_MS` |
//! | Cohere | [`CohereProvider`](cohere::CohereProvider) | `cohere` | `COHERE_API_KEY`, `COHERE_DEFAULT_MODEL`, `COHERE_MAX_RETRIES`, `COHERE_INITIAL_RETRY_DELAY_MS`, `COHERE_MAX_RETRY_DELAY_MS` |
//! | Groq | [`GroqProvider`](groq::GroqProvider) | `groq` | `GROQ_API_KEY`, `GROQ_DEFAULT_MODEL`, `GROQ_MAX_RETRIES`, `GROQ_INITIAL_RETRY_DELAY_MS`, `GROQ_MAX_RETRY_DELAY_MS` |
//! | `OpenRouter` | [`OpenRouterProvider`](openrouter::OpenRouterProvider) | `openrouter` | `OPENROUTER_API_KEY`, `OPENROUTER_DEFAULT_MODEL`, `OPENROUTER_SITE_URL`, `OPENROUTER_APP_TITLE`, `OPENROUTER_MAX_RETRIES`, `OPENROUTER_INITIAL_RETRY_DELAY_MS`, `OPENROUTER_MAX_RETRY_DELAY_MS` |
//! | OpenAI-compatible (Ollama, vLLM, `LocalAI`, …) | [`OpenAiCompatibleProvider`](openai_compatible::OpenAiCompatibleProvider) | `ollama` / `vllm` / `localai` / `local` | `LOCAL_LLM_BASE_URL`, `LOCAL_LLM_MODEL`, `LOCAL_LLM_API_KEY` |
//!
//! What they share lives in [`client`] (HTTP client construction, the one
//! status-to-[`RunnerError`](crate::types::RunnerError) mapping, the retry
//! policy) and [`sse`] (the line-buffering Server-Sent Events parser). What
//! they do not share — Gemini's `system_instruction` hoisting and
//! thinking-only retries, Cohere's typed event envelope and its rule that a
//! content-less message sinks the whole request, `OpenRouter`'s ranking
//! headers — stays in each provider's own module.
//!
//! Every provider accepts a caller-owned [`reqwest::Client`] through
//! `with_client`, so a host with one connection pool can hand it to all of
//! them. Tool calling rides on [`ChatRequest::tools`](crate::types::ChatRequest::tools)
//! and comes back on [`ChatResponse::tool_calls`](crate::types::ChatResponse::tool_calls).
//!
//! No request URL is ever written to a log by this module: Gemini's API key
//! travels in the query string, so a URL in a log line is a key in a log line.

/// HTTP client construction, status-code mapping and the retry policy.
pub mod client;
/// Cohere v2 chat provider.
pub mod cohere;
mod cohere_errors;
/// Google Gemini provider.
pub mod gemini;
/// Groq provider (OpenAI-shaped API on LPU inference).
pub mod groq;
/// Any OpenAI-compatible endpoint: Ollama, vLLM, `LocalAI`, and others.
pub mod openai_compatible;
/// `OpenRouter` provider (one key, many upstream models).
pub mod openrouter;
/// Line-buffering Server-Sent Events parser shared by every streaming provider.
pub mod sse;

pub use cohere::{CohereConfig, CohereProvider};
pub use gemini::{GeminiConfig, GeminiProvider};
pub use groq::{GroqConfig, GroqProvider};
pub use openai_compatible::{OpenAiCompatibleConfig, OpenAiCompatibleProvider};
pub use openrouter::{OpenRouterConfig, OpenRouterProvider};