• Why AGIFlow?
• Getting Started
• Key Features
• Contribution
• License
• Support

If you are developing a complex LLM workflow, you might have some difficulties with:

• Understand how output from one LLM affect the other LLM inference
• Disect LLM execution dependencies to improve performance
• Communicate with QA and non-technical users on how the workflow works
  ...and more

AGIFlow solves these issues by visualising LLM workflow automatically for you based on telemetry data. With a few integration lines of code, you can get AGIFlow up and running from development environment to production. We understand the challenges of bringing your LLM app to production and want to ensure you have a scalable, trustworthy infrastructure to do so.

The simpliest way to get start is to use AGIFlow's control-plane dashboard. If you wish to keep your workflow and data private, you can also run Agiflow app locally using docker compose and start tracing your application by:

1. Clone the Repository:

git clone https://github.com/AgiFlow/agiflow-sdks.git
cd agiflow-sdks

2. Run the development app:

cd dockers/dev
docker-compose up

3. Integrate with your application:

• Go to localhost:3000 to get the API key and follow this documentation to get telemetry from your LLM app to Agiflow dashboard.

• With docker compose development setup, the AGIFLOW_BASE_URL or api_endpoint option is http://localhost:3000/api/dataplane; same url applied for frontend analytics sdk's endpoint.

This mono-repo is designed for fullstack developers and product teams to develop scalable LLM product. To learn more about why and how we use mono-repo, please visit Nx, PNPM and Poetry Mono-Repo documentation.

We're making sources available on a rolling basis from our internal repo. For comprehensive documentation, please visit AGIFlow's docs. Here is an overview of our packages:

• dockers: Docker compose to run and self-host AGIFlow.
• apps/agiflow-controlplane: The AGIFlow's controlplane dashboard built with Vitejs React
• llm-mocks: Mocking library for LLM providers, simplifying TDD, reducing CI/CD costs on regression testing, and facilitating API development.
• agiflow-eval: Python SDK for LLM evaluations, supporting custom templates and multiple models.
• agiflow-sdk: Python SDK built on top of Open-Telemetry to collect LLM metrics, supporting prompt and model registry synchronization for multiple LLM usages.
• @agiflow/js-sdk: Web SDK for frontend analytics on how users interact with LLM apps, allowing full-stack traceability.
• @agiflow/web-feedback: Feedback widget with session replay and workflow visualization to get high-quality feedback from simple chat apps to complex agentic workflows.

We welcome contributions from the community! If you would like to contribute, please follow this guide.

This repository is a mono-repo that contains various packages, modules, and applications, each of which may be governed by different licenses. Unless otherwise explicitly stated within the respective package, module, or application subdirectory, the code in this repository is licensed under the Business Source License (BSL). For more details, refer to the License Notice.

If you have any questions or need support, please open an issue in this repository or contact our support team at [email protected].

To stay updated, get support, and engage with fellow developers, join our Discord community. Click the link below to connect with us and be part of the conversation!

Thank you for using AGIFlow SDKs! We look forward to seeing the amazing things you build.