Skip to main content
naga.ac/playground runs requests against the same models as the API, so you can test a prompt or compare models before you write code. Sign in to naga.ac first. The sidebar keeps each conversation, and you can return to it later.

Pick a mode and a model

The Playground has a mode for every kind of model, listed below. Pick one at the top of a new room, or pick any model in the model picker. A new room offers a few example prompts, and one click puts the text into the input. The price of the selected model shows next to the picker and links to its pricing. Model pages on naga.ac/models also have a Try in Playground link that opens the model in the Playground. On a wide screen, chat, speech, transcription and embeddings parameters sit in a panel on the right. On a narrow one they sit behind the settings button. Image and video settings sit in the input, next to the prompt.
Playground in chat mode, with the model picker and a conversation

Translate speech

A transcription model that also translates shows a Task switch in its settings. Transcribe keeps the language of the audio. Translate to English returns English text through /v1/audio/translations. The room remembers the task. A model that only transcribes has no switch.

Embeddings, moderation and rerank

These modes answer with numbers rather than text, so the Playground reads them for you. In Embeddings, put one text per line. Each text gets its vector, and with two or more lines a table shows the cosine similarity of every pair, with the closest pair marked. Copy vectors copies them as JSON. A model that lets you shorten its vectors shows a Dimensions field in the settings. In Moderation, the answer says whether the text was flagged and lists each category’s score from highest to lowest. In Rerank, type the query and one document per line. The answer lists the documents from most to least relevant with their scores. Each answer shows how long it took and, when the model has a price, what it cost.

Generate video

A clip takes from half a minute to several minutes, so the Playground renders it in the background. You choose the length, resolution, aspect ratio and sound from the values the model supports. The price of the clip shows before you start it. You can start several clips at once. A clip keeps rendering when you leave the page, and it appears in its room when you come back. NagaAI keeps the video file for seven days. Finished clips also stay in your browser, so you can play and download them after that.

Compare models

Select Compare next to the model picker to send one request to two to four models of the same kind at once. Pick the kind in the comparison’s header: chat, images, video, speech, transcription, embeddings, moderation or rerank. Each kind takes its own input, such as a text to speak, an audio file to transcribe, or a query and documents to rank. Each answer shows how long it took and what it cost at the model’s list price. Chat models keep their own conversation over several prompts. A clip from each video model renders in its own column, and you can play the audio of each speech model there too. Side by side puts each model in its own column. List puts the answers one under another, which is easier to read for long answers. On a phone a comparison opens as a list, so every model’s answer fits the screen. Switch to Side by side there to swipe between the columns. The first prompt saves the comparison to the sidebar, named after that prompt, and it opens again with its models, answers and layout. New comparison starts over with the same models and keeps the last one in the sidebar. An answer still streaming when you close the page is marked as stopped when you come back. Send the prompt again to get it.

Copy the request as code

Select API next to the model picker to see the current request as Python and cURL code. For chat it opens on the Responses API, with a Chat Completions tab next to it. Embeddings, moderation and translation use the OpenAI SDK, and rerank uses a plain HTTP request. Paste the code into your project, and set NAGA_API_KEY to your own key.

Billing

NagaAI charges Playground requests to your balance like API requests. They are made in the workspace picked in the switcher card: they appear in its Logs under the key name Playground and spend its budget. Once that budget is spent, the Playground answers with the same 403 credit_limit_exceeded as the workspace’s keys. It can call only the models the organization and that workspace allow.