Test the endpoint before you trust the model.

curlgeco is an AI playground for Hugging Face and OpenAI-compatible chat endpoints. Register the endpoints you actually use, stream a real answer out of each one, run the same prompts across all of them, and keep the logs. It is open source under Apache-2.0 and it runs with no backend of ours in the middle.

Open curlgeco Read the source

What it does

Six things, done properly.

curlgeco is the workbench we wanted while picking models for our own products, so it is built for the way that work actually goes: one place for the endpoints, one place for the results, and a log you can run again tomorrow.

Endpoint manager

Register the Hugging Face router, a dedicated Hugging Face endpoint, or anything that speaks OpenAI's /v1/chat/completions, then test the connection before you rely on it.

Streaming playground

A chat surface that streams tokens as they arrive and renders markdown, so you read the answer in the shape the model actually returns it.

Batch tests and compare mode

Run a suite of prompts across endpoints and read the results side by side, instead of keeping three browser tabs and a guess.

Replayable request logs

Every call is kept as a log you can open and run again, which is the difference between remembering a result and reproducing it.

Local-first, optional accounts

The workspace lives in your browser. Supabase signup and login are there if you want them, and the app works with them switched off.

Import, export, theme

Take the workspace with you between machines, and switch the theme. Your configuration is a file, not an account you cannot leave.

How it is built

No backend, and nothing to trust us with.

curlgeco is a static export. Provider calls go from your browser straight to the endpoint you configured, with no server of ours in between, which also means the target endpoint has to allow browser requests. Endpoint API keys stay in browser storage and are sent only to the provider you pointed them at.

Licence: Apache-2.0
Source: github.com/ugeco-opensource/curlgeco
Runtime: Next.js App Router, static export
Backend: None. Calls go browser to provider
Deploy: npm, Docker Compose, Cloudflare Pages, Helm on AKS
Design docs: HLD, LLD, OpenAPI, and schema, in the repo

Run it yourself

Clone it, or use the hosted build.

The repository carries the app, the brand pack, and the design artifacts: HLD, LLD, OpenAPI, and the Supabase schema. Contribution is the point of this line of work, so the design documents are public alongside the code.

$ git clone https://github.com/ugeco-opensource/curlgeco.git
$ cd curlgeco/submodule/curlgeco_frontend
$ cp .example.env .env
$ npm install && npm run dev

Docker Compose builds the same static export and serves it with nginx. For a hosted copy, Cloudflare Pages takes the repo as is, and the build we run sits at curlgeco.ugeco.in.

Open curlgeco

Bring your own endpoint.

Nothing to install if you do not want to. Point it at an endpoint and read a real answer back.

Open curlgeco Star the repo

Built by ugeco · ugeco open source