The AI framework that adds the engineering to prompt engineering (Python/TS/Ruby/Java/C#/Rust/Go compatible)

From a goal to a task DAG, automatically. TypeScript-native multi-agent orchestration with MCP and live tracing. Three runtime dependencies.

Repair malformed JSON from LLMs, APIs, logs, and user input in Python.

Schema-Guided Reasoning (SGR) has agentic system design created by neuraldeep community

MLX Omni Server is a local inference server powered by Apple's MLX framework, specifically designed for Apple Silicon (M-series) chips. It implements OpenAI-compatible API endpoints, enabling seamless integration with existing OpenAI SDK clients while leveraging the power of local ML inference.

Unified Go interface for Language Model (LLM) providers. Simplifies LLM integration with flexible prompt management and common task functions.

本地监控 + AI 视觉 — LAN-based smartphone-powered AI monitoring framework with structured event output for data acquisition and analysis.

A versatile workflow automation platform to create, organize, and execute AI workflows, from a single LLM to complex AI-driven workflows.

React Native Apple LLM plugin using Foundation Models

Simplifies the retrieval, extraction, and training of structured data from various unstructured sources.

OpenAPI definitions, converters and LLM function calling schema composer.

Where conversations become interfaces.

Dev tools, optimized for agents. Structured, token-efficient MCP servers for git, test runners, npm, Docker, and more.

Deterministic research MCP server on FastMCP 3 — 5-engine web search, 9-platform social search, 6 academic DBs, news aggregation, entity profiles, conflict detection, document analysis. No API keys. No in-server LLM. Structured outputs for agent chaining.

Structured output generation in Swift

🚬 cigs are chainable Ai functions for typescript. Call functions with natural language and get a response back in a specified structure. Uses OpenAI's latest Structured Outputs.

Runtime patches for vLLM — Qwen3.6 (27B int4 / 35B-A3B FP8) on consumer Ampere. 50+ patches: TurboQuant KV, MTP / DFlash / ngram spec-decode, FULL cudagraph, 256K-320K context. v7.64: P67 non-pow-2 GQA, Cliff 1 fix, 6 docs (FAQ/HARDWARE/CONFIGS/CLIFFS), Genesis Compat Layer.

Open-source agentic schema CLI. Optimised for claude code, gemini, codex and co-pilot. Skills included.

LLM-powered security log analyzer: detect threats & anomalies with zero regex — just declare a Pydantic schema. Real-time Telegram alerts, SIEM-ready with Elasticsearch/Kibana. Supports OpenAI, Ollama, vLLM.

[ti]ny [li]ttle machine learning [tool]box - Machine learning, anomaly detection, one-class classification, and structured output prediction