Connect OpenRouter

Experimental: read this first

Before you start

You’ll need Flashbag installed and activated through Flashbag Launcher, an OpenRouter account, and credit for any paid models you want to use. OpenRouter usage is billed separately from Flashbag and from your Claude or ChatGPT subscription.

1. Enable Pi in Flashbag

1. Turn on experimental Pi support

   Open Settings → Experimental and switch on Pi · advanced testing. This makes Pi’s settings available.

2. Enable or install Pi

   Go to the Agents tab within Settings. In the CLIs section, find pi and turn on Enabled. If Pi isn’t installed, use Install and let it finish.

3. Find OpenRouter sign-in

   Return to the Dashboard. Under CLI dependencies, find Pi and choose Sign in · openrouter. You can also open this from Pi’s settings. The connection asks for an OpenRouter API key.

2. Create your OpenRouter API key

1. Open API Keys

   Sign in to OpenRouter and open API Keys. Create a new key and give it a recognizable name, such as Flashbag.

2. Set a spending limit and expiration

   Choose a Credit limit that fits your budget. For a daily allowance, set Reset limit every… to daily. You can also choose an expiration date for the key.

   A daily limit renews each day; it is not a one-time total budget. Choose a limit without a reset if you want to cap the key’s total spending.

3. Create and copy the key

   Complete any verification OpenRouter requests. In the video, this includes entering a code sent by email. Copy the newly created key so you can paste it into Flashbag.

3. Connect the key to Flashbag

1. Paste your API key

   Return to the Connect OpenRouter to Pi dialog in Flashbag. Paste the key into the API key field and submit it to connect.

2. Check the connection

   Wait for Connected to Pi, then close the dialog. The OpenRouter entry under Pi should now show as signed in. Use Refresh in CLI dependencies if the Dashboard hasn’t updated yet.

4. Choose an OpenRouter model in chat

1. Open the model picker

   Open a chat with your agent and select the model control beneath the message box.

2. Switch to Pi

   Choose the Pi tab, then search or browse the OpenRouter models. Select the model you want to try.

3. Send a short message

   Start with a simple request and check the reply. You can return to the picker whenever you want to use a different model.

Changing the AI tool or model starts a new session. Your agent’s memory and recent conversation are preserved, as the model picker explains.

No web search out of the box

Agents using this integration currently have no built-in web search. They cannot look up current information or check their answers against web sources out of the box. They rely on the model’s internal knowledge, the context you provide, and whatever files or tools they can access locally.

That knowledge can be outdated or incorrect, even when an answer sounds confident. Verify important claims independently, especially when a task depends on recent information.

Keep an eye on usage costs

In the walkthrough, Mitchel sends the same “hello” to two models and gets very different costs. Your message is only part of the request: the agent’s instructions and loaded memory also contribute to the input.

Use a cost-effective model for everyday work and check usage before moving to a more expensive one. Review your balance and spending in OpenRouter. The amounts shown in the video are examples from those requests, not a fixed price per message.

If something doesn’t connect