$ show microsoft/TypeChat
TypeChat is a library that enables building natural language interfaces by leveraging type definitions (TypeScript interfaces, Zod schemas, Python TypedDicts/dataclasses) to structure LLM interactions. Instead of prompt engineering, it uses schema engineering: types define the expected output structure, which is used to construct prompts, validate LLM responses, and repair invalid outputs through iterative LLM calls. The library provides a clean abstraction over LLM APIs (OpenAI/Azure OpenAI) with retry, timeout, proxy, and response-size limiting. It supports both direct JSON translation and program generation (multi-step function calls). Implemented in TypeScript and Python with parallel APIs, it includes a rich set of examples (sentiment analysis, coffee shop ordering, calendar management, music control, math, drawing, health data, crossword, multi-agent routing) and comprehensive test suites. The codebase demonstrates strong engineering practices: clean separation of concerns (model, validator, translator), well-tested schema conversion (Python types to TypeScript declarations), and safe program interpretation. While still in early version (0.1.4), it is actively maintained with CI/CD, dependabot alerts, and a documentation site.
The LLM identified 4 files worth highlighting for reuse:
Provides a production-grade LLM client abstraction (TypeChatLanguageModel interface) with configurable retry logic (max attempts, pause, Retry-After header support), request timeout via AbortSignal, response size limiting via streaming body reader, and proxy support via undici (EnvHttpProxyAgent, ProxyAgent). It supports both Chat Completions and Responses API. This module is self-contained (only depends on standard Node.js and optional undici) and could be extracted as a standalone LLM client library.
Converts Python type annotations (TypedDict, dataclass, generic alias, tuple, union, Literal, etc.) into an internal TypeScript AST node representation. Handles complex cases like conflicting names, generic aliases, abstract collections, and type alias syntax. The output is then serialized to TypeScript declaration strings via ts_node_to_string.py. This module is well-tested across multiple Python versions and could be reused in any project needing to generate TypeScript type definitions from Python types.
Implements a safe JSON program interpreter (evaluateJsonProgram) that executes a sequence of function calls defined in a JSON structure. It validates function names against a whitelist, checks argument references, and dispatches calls via a user-provided callback. Includes security measures: prototype-member blacklist, bounds checking on result references, and injection-safe function name validation. This pattern is reusable for any scenario requiring controlled execution of LLM-generated multi-step plans.
Provides a reusable REPL loop (processRequests) that reads user input from stdin or a file, processes each line through a provided callback, and prints results. Supports quit/exit commands and optional file input. This utility is shared across all TypeScript examples and could be extracted for any CLI application needing an interactive prompt loop.