briglia menu: setup and settings

Pick how Briglia thinks

Right after choosing the language, the page asks which AI Briglia should use:

• OpenRouter (recommended, selected by default on a new setup). New setups start on DeepSeek V4.1 Flash (deepseek/deepseek-v4.1-flash), pinned to the Together host, the fastest one when we tested; you can change the model later here or with /model, and the host with /orprovider. If OpenRouter can't confirm that Together serves the model, nothing is pinned, the page tells you, and OpenRouter picks the host. Your OpenRouter key covers everything else too: reading web pages, voice messages, image creation (Google's Gemini image models) and scanned documents. No OpenAI key needed.
• ChatGPT subscription. Needs ChatGPT Plus or Pro. Nothing is billed per message: usage counts toward your normal ChatGPT limits, and web research runs through the subscription too.
• OpenCode Go, or a server: a model on this computer or your network (LM Studio, Ollama, vLLM) or any other OpenAI-compatible provider with its API key. You give each server a name and can add as many as you like. These lanes also need an OpenAI API key, because page reading runs on OpenAI models (the only ones fast enough for it).

Since CLI 0.2.54 OpenRouter is the recommendation for new setups (before, ChatGPT was selected first). An existing setup keeps the provider and host pin it already has.

The thinking level defaults to Deep (high) on every provider. The page shows friendly names with the real level underneath: Light (low), Balanced (medium), Deep (high), Deepest (xhigh).

What else you need

• Telegram on your phone. That's where you talk to Briglia.
• Free accounts at serper.dev (Google search) and jina.ai (reading web pages). The page tells you where to click.
• On the ChatGPT lane, the OpenAI API key is optional: it's needed for voice messages and image creation (separate from ChatGPT, billed per use by OpenAI). On the OpenRouter lane it's optional too: voice and images already work through OpenRouter, and the page shows them as Via OpenRouter; if you add an OpenAI key, Briglia uses OpenAI for them instead. Also optional: an AgentMail key, which gives Briglia its own email address and calendar.
• Mac: Homebrew, for the document and media tools.

1. Install Briglia

   Open the Terminal app. Mac: press ⌘ + Space, type Terminal, press Enter. Linux: Ctrl + Alt + T. Paste this and press Enter:

   curl -fsSL https://github.com/permaevidence/briglia-cli/releases/latest/download/install.sh | bash

   Copy

2. Open the menu

   briglia menu

   Copy

   Close the Terminal window, open a new one (Mac: ⌘ + N) so it finds the newly installed briglia, then type the command and press Enter. Your browser opens the Briglia page. If it doesn't, the terminal shows a link to open on the same computer. Choose English or Italiano; the flags in the top corner switch language any time. Everything is saved as you go, so you can close the page and come back later. Keep the terminal window open while you use it.

3. Connect the AI

   ChatGPT: click Sign in with ChatGPT and approve Briglia in the tab that opens. Nothing else needs enabling. For a remote setup, choose Use a code and type the code on any device. That option needs one ChatGPT setting: Settings → Security and Login → “Enable device code authorization for Codex”. Briglia starts on GPT-6.1 Sol (since CLI 0.2.44; an existing setup keeps its model).

   OpenRouter: paste your OpenRouter key and pick a model. That's the only AI key it needs.

   OpenCode Go: paste your OpenCode key, pick a model, and add the OpenAI key.

   A server: type its address, add its API key only if it asks for one, press Find models, pick a model, give the server a name (for example Home GPU), and add the OpenAI key. A key is only ever sent over https, or over plain http to this computer or your home network.

4. Connect Telegram

   In Telegram, open @BotFather, send /newbot, pick a name, and copy the token it gives you. Paste it in the page. Then send your new bot any message: the page finds you by itself and asks “Is this you?”. You don't need to look up a chat ID (you can still type it if you prefer).

5. Paste the keys

   Serper, then Jina, then the optional ones. Each key is checked and saved the moment you paste it; there is no verify button. If a key is refused, the page says why in plain words and you paste it again.

6. Let it set up the computer

   • Mac: Full Disk Access for your terminal app. The page opens the right System Settings pane and notices by itself when it's on. If macOS asks to quit and reopen the terminal, do it and type briglia menu again. Keep-awake is automatic while Briglia runs.
   • Linux: the page checks that the computer won't suspend by itself and offers the fix.
   • Document & media tools (PDF, Word, Excel, audio, video): one choice installs whatever is missing. The first time it can take 10–20 minutes, and on Linux you may be asked for your password.

7. Start Briglia

   Mac: Briglia runs in that terminal window. Keep it open (minimized is fine). Linux: it runs in the background and starts by itself when the computer turns on. The page only says it's running once Briglia really answers. Now write to your bot on Telegram.

Changing things later

Type briglia menu again. The page opens on Your AI lanes: one card for every AI you've set up (ChatGPT, OpenCode Go, OpenRouter, each named server, and the OpenAI API if you set it up with briglia setup), showing its model and thinking level. The one Briglia thinks with is marked In use.

At the top, Briglia is using (Briglia sta usando) shows the lane in use. Tap it to open the list of every lane you've set up and choose another: Briglia switches right away. The others stay saved with their own model, key and thinking level. If Briglia is in the middle of an answer, the page asks you to try again once it has replied; if the lane was changed from Telegram meanwhile, the page refreshes and says so instead of switching.

• Edit opens the lane: change its model or thinking level, replace a key, sign in to ChatGPT again, or, for a server, rename it, change its address or key, or pick another model. Editing the lane in use applies from the next message.
• Remove asks once, then deletes the lane (for ChatGPT: signs out). The lane in use can't be removed; switch to another one first.
• + Add a lane adds ChatGPT, OpenCode Go or OpenRouter (one of each) or another server (up to 20). Adding a lane saves it and puts it in the Briglia is using list; it doesn't switch. The only exception is the very first lane of a new setup, which is used straight away.

Below the lanes, Other settings has everything else:

• Telegram, keys, email: connect a different bot, replace or remove a key, turn email on or off.
• This computer: permissions, keep-awake, tools.

Once setup is complete, a Start/Stop switch sits in the top-right corner of every screen: green Running · Stop while Briglia runs, grey Stopped · Start otherwise. Stop asks once before it acts. On Linux, Stop also turns off start at boot.

Switching from Telegram

Send /provider to Briglia: every lane you've set up is a button, and each server appears under its own name. You can also type /provider Home GPU. If you switch to OpenCode Go or a server while no OpenAI key is saved, Briglia switches but warns that web research will fail until you add the key. /model and /effort then change the lane in use.

Other ways to set up

Quick setup (briglia quicksetup) is the older one-page form and stays available as a fallback. The step-by-step wizard (briglia setup) asks the same questions in the terminal. All three save to the same place, so you can mix them.