Use Obscurify.ai in Brave Leo

Brave’s built-in assistant, Leo, can run on a model you bring yourself. Add Obscurify.ai’s OpenAI-compatible API under Bring your own model, and Leo’s chat and page questions run on any model your Obscurify.ai plan includes. Brave sends these requests straight from your browser to Obscurify.ai, not through Brave’s servers.

You need an Obscurify.ai account (a random account number, no email) and its API key: create an account, then click Show API Key on the account page. The API works on every plan with a positive balance. New accounts may get $1 of starter credit while this month’s promo budget lasts.

Add a model

  1. Open brave://settings/leo-ai (Settings → Leo).
  2. Under Bring your own model, click Add new model.

Bring your own model section in Brave’s Leo settings

  1. Fill in the form:
    • Label: any name, for example Obscurify.ai DeepSeek V4 Flash.
    • Model request name: a model ID from the model list, for example deepseek/deepseek-v4-flash. It has to match exactly.
    • Server endpoint: https://obscurify.ai/v1/chat/completions (the full path, not just /v1).
    • Context size: how many tokens Leo may send. The default is 4000; raise it so Leo can send more of a long page or conversation. We used 64000.
    • API Key: your sk-obs1_... key.
    • Leave Vision Support and Tool Support off.
    • System Prompt: optional. Leo sends its own short system prompt (the grey text in the box) with every request.
  2. Click Add model. Repeat for each model you want in Leo’s menu.

Brave’s Add model form filled in for Obscurify.ai

Chat

  1. Open Leo from the sidebar, or go to brave://leo-ai.
  2. Click the model name under the message box, then Show all models. Your Obscurify.ai models are at the bottom. Brave marks every model you add yourself as Local, even when it runs on a server.

Obscurify.ai models in Leo’s model menu

  1. Ask away. To ask about a page, click the paperclip, choose Open tabs… and pick the tab.

Leo summarizing an attached page with an Obscurify.ai model

What works

We tested deepseek/deepseek-v4-flash, meta-llama/llama-3.3-70b-instruct, anthropic/claude-haiku-4.5, openai/gpt-5-mini, google/gemini-3.8-flash, deepseek/deepseek-v4-pro-0813:thinking and mistralai/mistral-medium-3.1.

  • Streaming: yes.
  • Attached tabs: yes. Leo sends the page text along with your question.
  • Reasoning models: yes. Leo shows the answer, not the reasoning.
  • Conversation titles: Leo asks the model to title each chat. Some models return clumsy titles; it doesn’t affect the answers.
  • Errors: Leo shows “Something went wrong” for every error, whether the key is wrong, the balance is empty or the model isn’t on your plan. To see the real message, send the same key and model with curl.

Plans

Prices per model are on the model list, and each model has its own page with its price and the plan it needs. Frontier models need the Casual plan; uncensored models need Pro. See pricing.

Support

Questions or problems: contact us (email optional).

Last checked with Brave 1.98.60 on 2026-10-28.