nexu-io/open-design
Open Design is a local-first, open-source desktop application that serves as an alternative to Claude Design, integrating with various coding agents to generate web, desktop, and mobile prototypes,...
Awesome AI Agents › Generative Software Frameworks
generative-computing/mellea
Mellea is a Python library for writing what it calls generative programs, replacing free-form prompts and unpredictable agent loops with structured, testable workflows around large language model calls. Its central mechanism is a decorator that turns an ordinary type-annotated Python function into a structured model call: the docstring becomes the prompt and the type hints become the output schema, so no prompt templates and no output parsers are written by hand. A Pydantic model used as the return annotation is enforced at generation time, which means fields come back with the declared types rather than as text that must be validated afterwards. Calls run inside a session object, obtained from a convenience entry point that supplies sensible defaults which can be overridden. On top of structured output the library adds requirements and repair: natural-language requirements can be attached to any call, and Mellea validates the result against them and retries automatically when they are not met. Sampling strategies let the same generation be run several times and the best result selected, switching between rejection sampling, majority voting and other strategies by changing a single parameter. Several backends are supported, covering local inference through Ollama, hosted APIs from OpenAI and IBM WatsonX, HuggingFace models, LiteLLM and Bedrock. A helper is provided to drop the library into existing codebases incrementally rather than rewriting them, and any generative program can be exposed as a Model Context Protocol tool so other agents can call it. Documentation includes tutorials, an API reference, Colab notebooks and runnable examples covering retrieval-augmented generation, agents, the instruct-validate-repair pattern and the library object model. The project was started by IBM Research and is released under the Apache 2.0 license.
https://github.com/generative-computing/mellea
Also in Generative Software Frameworks
deepset-ai/haystack
Haystack is an AI orchestration framework for building customizable, production-ready NLP applications using large language models, vector search, and advanced retrieval techniques.
op7418/guizang-ppt-skill
AI agent skill for generating polished HTML slide decks with various design styles, image prompts, social covers, and a WebGL/low-power presentation runtime.
HKUDS/Agentic-AIGC
ViMax is an agentic AI system for end-to-end video generation, transforming raw ideas, novels, or scripts into complete video content by orchestrating scriptwriting, storyboarding, character creati...
The-Pocket/PocketFlow
Pocket Flow is a minimalist 100-line LLM framework that uses a graph-based abstraction to enable expressive multi-agent systems, workflows, and retrieval-augmented generation, designed for high productivity and minimal bloat.
Netflix/metaflow
Metaflow is a human-centric framework by Netflix that streamlines building, managing, and deploying real-life AI and machine learning systems from prototyping to production at scale.
presenton/presenton
Presenton is an open-source AI-powered presentation generator and API that runs locally, offering customizable layouts, multiple AI model integrations, versatile image generation, and privacy-focused presentation creation and export.
thesysdev/openui
OpenUI is a full-stack Generative UI framework that offers a compact, streaming-first language for model-generated UI, a React runtime, and token-efficient chat interfaces.