Don't Tattoo the Model Name on Your Demo Yet: Gofannon Lets Agent Prototypes Get the User Journey Working First
When building Agent prototypes, the most common mistake is binding to a model provider first. Gofannon uses a guided interface to assemble tools, data sources, and decision paths, preview in real time, and choose the model last. Open source, Apache 2.0.
Gofannon is an open-source toolkit and web app with a very specific goal: let builders and business experts first build a working Agent prototype, then decide which model provider to use. It is maintained by The AI Alliance and licensed under Apache 2.0.
People building Agent demos often have the impulse to pick the model first and then build the interface around it. Gofannon does the opposite, putting the model provider later. It provides a guided interface that lets users combine tools, data sources, and decision paths, then preview Agent interactions in real time. The lightweight web UI can connect forms, chat interfaces, and dashboards to validate the complete user journey.
This solves a very practical problem: once the model changes, the demo's storyline, tool calls, and interface feedback may all change. Getting the flow working early reduces rework later when choosing a provider. In replies, The SI Maximalist said that testing the user journey first and binding to a model provider later is a sensible order of operations. Hongji Lee made an analogy: keeping a demo model-agnostic is like not tattooing your current partner's name on your body. E1ev3n also mentioned having seen demos fall apart as soon as the model was changed, so they would rather see how users actually use it first and then pick the right provider.
Functionally, Gofannon supports multiple model providers, including OpenAI, Anthropic, Gemini, and others, connected through LiteLLM. The project remains provider- and model-agnostic, making the prototypes built with it easier to migrate.
Quick start is simple:
git clone https://github.com/The-AI-Alliance/gofannon.git
cd gofannon/webapp/infra/docker
docker-compose up --build
Environment configuration and details are in the quickstart guide. Full documentation is published here, including API reference, LLM provider configuration, and testing instructions. The repository is at The-AI-Alliance/gofannon.
The name has an origin too. Gofannon is the god of smithing in Welsh mythology. What the project wants to do is similar: put tools, data, and decision paths on a workbench, first forge a usable prototype, and only then talk about what materials to swap in later.
For developers, its value lies in decoupling Agent prototypes from model providers; for business experts, the key is being able to assemble the user journey directly in the interface and see feedback in real time. Don't rush to lock in a provider during the prototype stage; this order is more stable.
Original tweet link: Dan Kornas's introduction.
发布时间: 2026-11-01 07:33