Features & Optional Modules

This guide explains how to include ai-lib in your Cargo project with opt-in features, along with recommended presets for different scenarios.

Basic Dependency

[dependencies]
ai-lib = "0.3.4"
tokio = { version = "1", features = ["full"] }

Enabling Features

ai-lib ships lean-by-default. Turn on features as needed:

[dependencies]
ai-lib = { version = "0.3.4", features = ["resilience", "streaming"] }

Feature Aliases (for ergonomics)

resilience → enables interceptors (retry, rate-limit, circuit-breaker wiring)
streaming → enables unified_sse (unified streaming parser)
transport → enables unified_transport (shared reqwest client factory)
hot_reload → enables config_hot_reload (config provider/watch traits)
all → turns on most OSS features: interceptors, unified_transport, unified_sse, cost_metrics, routing_mvp, observability, config_hot_reload

These aliases are additive; they do not add new code, only enable existing granular features.

New: Basic Failover (OSS)

Since v0.3.x, AiClient::with_failover(Vec<Provider>) enables automatic provider switching on retryable errors (network, timeout, rate limit, 5xx). When used with routing_mvp, model selection is preserved during failover.

Suggested Combos

Minimal app: no features (add only when needed)
Production app: resilience, transport, streaming
Advanced ops: resilience, transport, streaming, observability
Dynamic config: add hot_reload

Example: Production-lean Setup

[dependencies]
ai-lib = { version = "0.3.4", features = [
  "resilience",
  "transport",
  "streaming"
] }

Import Style

Prefer root imports for developer ergonomics:

// Recommended for applications
use ai_lib::prelude::*;

// Or explicit imports for library development
use ai_lib::{AiClient, Provider, ChatCompletionRequest, Message, Role, Content};
use ai_lib::{Tool, FunctionCallPolicy};

Notes

Keep your dependency surface minimal: enable only what you use.
Features are forward-compatible; we avoid breaking renames.
See also: Getting Started, Chat & Streaming.