WeChatAI

Abstracts OpenAI, Azure OpenAI, and GPT-3.5/GPT-4 endpoints behind a single Rust-based client interface, handling provider-specific authentication, request/response serialization, and error mapping. Routes requests to the appropriate provider based on configuration without requiring application-level provider detection logic.

Provides a templating system that supports variable substitution, conditional blocks, and dynamic prompt composition using a custom template syntax. Parses template strings at compile-time or runtime, validates variable references, and renders final prompts with user-supplied context dictionaries, enabling reusable prompt patterns without string concatenation.

Maintains and manages multi-turn conversation state by storing message history (user/assistant pairs) in memory, implementing sliding-window context management to respect token limits of underlying LLM models. Automatically truncates or summarizes older messages when conversation exceeds model-specific context windows, preserving recent exchanges for coherent multi-turn interactions.

Constructs properly-formatted chat completion requests for OpenAI and Azure OpenAI APIs by mapping application-level parameters (temperature, max_tokens, top_p) to provider-specific request schemas. Handles provider differences in parameter naming, validation ranges, and required fields, ensuring requests conform to each provider's API specification without manual schema translation.

Parses unstructured LLM text responses and extracts structured data (JSON, key-value pairs, markdown) using pattern matching and optional JSON schema validation. Handles malformed or partially-complete responses gracefully, attempting to extract valid data from incomplete or corrupted LLM outputs without failing the entire request.

Serializes conversation history and LLM responses to markdown format with proper formatting (code blocks, headers, emphasis), enabling human-readable export of chat sessions. Supports custom markdown templates for conversation structure, preserves formatting from LLM responses (code blocks, lists), and generates exportable markdown files suitable for documentation or archival.

Loads and manages application configuration (API keys, model names, provider endpoints) from environment variables, configuration files (TOML/YAML), or command-line arguments with a hierarchical override system. Validates configuration at startup, provides sensible defaults, and supports multiple configuration profiles for different deployment environments (dev, staging, production).

Implements comprehensive error handling for API failures, network timeouts, and rate limiting with automatic retry logic using exponential backoff. Distinguishes between retryable errors (rate limits, transient network failures) and non-retryable errors (authentication failures, invalid requests), applying appropriate retry strategies to each error class.

Provides structured logging for API requests, responses, and errors with configurable log levels and output formats. Logs request/response payloads (with optional PII redaction), timing information, and error details to enable debugging and monitoring of LLM interactions. Supports multiple log outputs (stdout, files, structured JSON) for integration with observability platforms.

Supports processing multiple chat requests concurrently using Rust's async/await runtime, enabling efficient batch operations on large conversation sets. Implements connection pooling and request queuing to manage concurrent API calls without overwhelming provider rate limits, with configurable concurrency limits and request batching strategies.