732124645/promptops

PromptOps is an open source runtime platform that pulls the prompts, agent configurations, and workflows used by AI applications out of the source code and puts them behind an API and a web UI. The pitch in the README is direct: stop hardcoding prompts in your application. Once a prompt lives in PromptOps, you can edit it, version it, publish it, gray release it (send a slice of traffic to a new version), and watch how it behaves, all without redeploying the app that uses it. The backend is written in Go using Gin, GORM, and SQLite, and listens on port 8080. The web UI is a Vue 3 single page app, and the team ships small SDKs in Node, Python, and Java so an application can fetch a prompt by a key like sql.generator, fill in template variables written as double curly braces, and subscribe to a WebSocket endpoint so the local cache updates the moment someone edits the prompt in the dashboard. The Playground page in the UI lets a user fill in the variables of a prompt, send it to a model provider (mock, OpenAI, Claude, Ollama, or Gemini), and compare results across versions. Agents bundle a prompt with a provider and model and can be run with one click. Workflows chain steps together, render then model then transform, and record a trace per step. Running it locally is meant to be quick: docker compose up build, then open http://localhost:8080 and log in with admin and admin. The README is explicit that the defaults, including a static bearer token called promptops-dev-token and the admin password, are not safe for the public internet. It points to SECURITY.md for the hardening checklist. Role based access splits users into admin, editor, and viewer.