Orchestration AI apps

Frontend stack for building in-app AI copilots and agents.

CopilotKit is a framework for embedding AI copilots and agents directly inside web apps — chat, generative UI that renders live React components, shared application state, and frontend tool calls. It connects to agent frameworks like LangGraph, CrewAI, LlamaIndex, and Pydantic AI rather than shipping its own agent, and offers both headless APIs and prebuilt UI components. The company says its stack is used across the majority of Fortune 500 companies and powers millions of agent-user interactions per week.

Open-source RAG engine with deep document understanding.

RAGFlow is an open-source retrieval-augmented generation engine that turns complex documents—PDFs, slides, spreadsheets, scans, and web pages—into grounded, citation-backed context for LLMs. Its DeepDoc parser and hybrid vector plus full-text search aim for reliable, hallucination-resistant question answering, and it now bundles an agent-orchestration layer. Self-host the Apache-2.0 engine or use the managed cloud.

Model-driven, open-source agent SDK for Python and TypeScript.

AWS's open-source SDK for building production agents with a model-driven approach: you supply a model, tools, and a prompt, and the model plans and drives the loop. Model-agnostic (Amazon Bedrock, Anthropic, OpenAI, Ollama, and more), with native MCP tool support, 20+ built-in tools, conversation/context management, hooks, guardrails, and built-in OpenTelemetry observability.

Barebones Python library for agents that think in code.

Hugging Face's minimal agent library: agents express their actions as executable Python code snippets rather than JSON tool calls, which tends to take fewer steps than tool-calling loops. Model-agnostic via LiteLLM (local Transformers, OpenAI, Anthropic, and 100+ providers), with sandboxed execution (E2B, Docker, Modal) and tools/agents shareable on the Hugging Face Hub.

The control plane for enterprise AI agents.

Credal is a platform for building, governing, and deploying enterprise AI agents and MCP servers powered by company data. It syncs and inherits permissions from 50+ source systems, redacts sensitive data, and logs every run for audit, so security and governance are woven into each agent. Teams ship no-code agents to surfaces like Slack, ChatGPT, and Claude while keeping centralized access controls and compliance.

Open-source, code-first framework for building and deploying multi-agent systems.

Agent Development Kit (ADK) is Google's open-source framework for building, evaluating, and deploying AI agents and multi-agent systems with a code-first, software-engineering workflow. Agents are defined in code and composed into hierarchies, equipped with prebuilt tools, MCP tools, third-party libraries, or other agents. ADK supports bidirectional audio and video streaming, ships built-in evaluation, and deploys anywhere — your own infrastructure, Cloud Run, GKE, or Vertex AI Agent Engine.

Enterprise iPaaS for integration, automation, and AI agents.

Workato is an enterprise integration and automation platform (iPaaS) with 1,200+ prebuilt connectors spanning CRMs, ERPs, HR systems, and databases. Its Workato ONE layer adds agentic orchestration — Otto, Agent Studio, and an enterprise MCP gateway — so teams build, run, and govern AI agents alongside traditional low-code recipe workflows.

Open-source framework for real-time voice and multimodal AI agents.

Pipecat is a Python framework for building voice and multimodal conversational agents that can listen, speak, and see in real time. It orchestrates streaming speech-to-text, an LLM, and text-to-speech into one low-latency pipeline, wiring together 40+ AI services with no vendor lock-in. Client SDKs cover JavaScript, React, React Native, Swift, Kotlin, C++, and ESP32, and Pipecat Cloud offers managed hosting at scale.

Build and deploy durable AI agents and background workflows in TypeScript.

Trigger.dev is an open-source framework for writing long-running background tasks, AI agents, and workflows as plain TypeScript — with no timeouts, automatic retries, durable cron schedules, queues, and built-in observability. It runs on a fully managed cloud or self-hosted via Docker or Kubernetes under Apache 2.0. Features include human-in-the-loop steps, streaming LLM responses, and Python and browser-automation tasks.

Open SDK for production AI agents in .NET and Python.

Microsoft's multi-language framework for building, orchestrating, and operating AI agents and multi-agent workflows. It offers a consistent API across C#/.NET and Python with first-class support for both, plus built-in durability, restartability, observability, and human-in-the-loop control. Provider-flexible, so the underlying models can change without rewrites.

Visual low-code builder for AI agents and RAG apps.

Open-source tool for prototyping and deploying agentic and RAG applications through a drag-and-drop canvas, with Python customization underneath each node. It connects major LLMs, vector databases, and tools, and can expose flows as APIs or MCP servers. Run it locally, as a desktop app, or on managed cloud.

Open-source Python framework for production RAG and agents.

Orchestration framework for building LLM applications as modular pipelines — retrieval, routing, memory, and generation wired together with explicit, traceable data flow. It is model- and store-agnostic, integrating major providers and vector databases behind a stable component API. Aimed at production: serialization, logging, and deployment across cloud or on-prem.

Orchestration for data, AI, and infrastructure workflows as declarative YAML.

Kestra is an open-source orchestration platform for building data, AI, and infrastructure pipelines as declarative YAML workflows. It is event-driven and language-agnostic, running Python, Bash, Node, Go, or containers across 1,500+ plugins, with a web UI for monitoring plus governance features like RBAC and audit logs. It ships as free self-hosted open source alongside paid Cloud and Enterprise editions.

High-performance Python framework for multi-agent systems.

Agno is an open-source Python framework for building multi-agent systems with memory, knowledge, tools, and reasoning. Agents are fast to instantiate and can be composed into teams, and the companion AgentOS runtime serves them in production as a FastAPI app with a control plane.

Program — don't prompt — your language models.

DSPy is an open-source Python framework for building LLM applications by writing modular code instead of brittle prompt strings. You declare input/output signatures and compose modules, then DSPy compiles and optimizes the underlying prompts — and can even tune model weights — for classifiers, RAG pipelines, and agent loops.

Open-source AI coding agent for VS Code, JetBrains, and the CLI.

Kilo Code is an open-source AI coding agent that plans, writes, edits, and runs code inside your editor. It offers specialized modes — Code, Architect, Ask, Debug, and Orchestrator — works across VS Code, JetBrains IDEs, and the CLI, and supports 500+ models with bring-your-own-key and zero markup. A free tier plus paid cloud and team options round it out.

Open-source visual builder to create, deploy, and orchestrate AI agents.

An open-source workspace for building AI agents on a Figma-like drag-and-drop canvas, conversationally, or in code. It connects LLMs to 1,000+ integrations plus knowledge bases and structured tables, then deploys a workflow as an API, scheduled job, webhook handler, or standalone chat app. Apache-2.0, YC-backed, and runnable in Sim's cloud or self-hosted via npm or Docker.

Build, evaluate, and deploy production LLM apps and agents.

An end-to-end development platform for building, testing, and shipping LLM applications and agents. Vellum pairs a visual drag-and-drop workflow builder with a Python SDK, and bundles prompt versioning, RAG, evaluation, and production monitoring in one place so technical and non-technical teammates can collaborate. Built-in eval and test suites let teams measure quality before and after deploy. A free tier is available; paid Pro and Enterprise plans add seats and scale.

TypeScript framework for building AI agents and workflows.

An open-source TypeScript stack for AI applications: agents, a graph-based workflow engine (.then/.branch/.parallel), memory, tools, and built-in observability behind one API. Run it self-hosted under Apache 2.0, or deploy to Mastra Cloud with a free Starter tier and paid Teams/Enterprise plans.

Type-safe Python agent framework, the Pydantic way.

An open-source Python framework for building production-grade agents with validated, structured outputs instead of raw-string parsing. Model-agnostic across OpenAI, Anthropic, Gemini and many more, with composable tools, durable execution, MCP support, and built-in observability via Pydantic Logfire. MIT-licensed from the team behind Pydantic.

Visually build AI agents and LLM workflows — drag-and-drop, self-hosted.

A low-code visual builder for AI agents, chatflows, and multi-agent systems on a node canvas. It self-hosts locally through npm or Docker, or deploys to major clouds, and is provider-agnostic across many LLMs via its node ecosystem. The core is Apache-2.0; a managed Flowise Cloud tier exists, making it freemium.

The data framework for LLM apps — RAG, agents, and document workflows.

An open-source framework (Python + TypeScript) for connecting LLMs to your data — ingestion, indexing, retrieval, and agentic document workflows. Pairs with the managed LlamaCloud (LlamaParse) for production parsing and extraction. The most-used RAG framework after LangChain.

One OpenAI-compatible API in front of models from every provider.

A unified gateway that routes a single endpoint and API key to models from Anthropic, OpenAI, Google, Meta, DeepSeek, xAI, and more — swap models by changing one parameter, with automatic fallbacks and one consolidated bill. Pass-through token pricing plus dozens of free models.

Visual platform for agentic workflows, RAG pipelines, and LLM apps.

An LLMOps platform that bundles a drag-and-drop workflow builder, RAG pipelines, agent tooling, model management, and observability into one surface — prototype to production without much glue code. Connects hundreds of proprietary and open models across providers. Self-host the source-available edition for free, or use Dify Cloud's paid tiers.

Multi-agent framework with explicit roles and tasks.

Python framework for orchestrating crews of specialised agents — researcher, writer, reviewer — coordinated through shared context. Opinionated about roles, sequencing, and delegation; good fit for content-and-research pipelines.

TypeScript SDK for streaming, tool-calling, and structured outputs.

Vercel's batteries-included TypeScript framework for LLM-powered apps. Streaming primitives, structured outputs, tool calling, and React hooks for chat UIs — works with every major provider out of the box.

Graph-based agent orchestration. Stateful loops with checkpoints.

LangChain's agent layer. Model agents as nodes in a state graph with persistent checkpoints, human-in-the-loop steps, and durable execution. Strong when you need a long-running, debuggable agent rather than a one-shot chain.

Durable workflow engine for AI background jobs.

Event-driven, durable execution engine — pause/resume, retries, fan-out, scheduling — designed for long-running AI jobs that can't sit in a request/response cycle. TypeScript-first; framework-agnostic.

The default open-source framework for composing LLM apps.

Python + TypeScript framework for chaining prompts, tools, retrievers, and memory into LLM applications. Ubiquitous in the ecosystem; pairs with LangGraph for agent orchestration and LangSmith for tracing.

Fair-code workflow automation with first-class AI nodes.

Visual workflow builder bridging APIs, databases, and AI providers. Self-hostable; commercial cloud available. The default Zapier-style surface for the agentic-workflow crowd.