BrainDrive Concierge Scoping Call (What it is & Help us Build!)

Help Build BrainDrive
1

Hi Guys,

Below is a recording of a call Dave J and I did today where we discussed building BrainDrive Concierge.

The BrainDrive Concierge will be the first AI model you meet when you install BrainDrive.

It’s designed to do two key things:

  1. Help you get up and running – Answer questions, explain how things work, and guide you through building, customizing, and benefiting from your own AI system.
  2. Showcase our core interaction – Chatting with your own AI system, right out of the box.

Unlike generic AI assistants, the Concierge is part of your BrainDrive. That means you own and control it, just like the rest of your system. No hidden services. No Big Tech behind the scenes. Just your AI, helping you make the most of your BrainDrive.

:hammer_and_wrench: How We’re Thinking About Building It

Here’s an overview of the current plan—this is all open to input and collaboration.

:speech_balloon: Core Functionality

  • Runs locally with BrainDrive.
  • Launches automatically with the chat interface.
  • Welcomes the user, introduces BrainDrive, and offers guidance.
  • Answers questions about ownership, setup, plugins, customization, and monetization.
  • Gives helpful links (GitHub, forum, docs) based on the type of question asked.
  • If it can’t help, it directs the user to the support forum.

:brick: Architecture Approach

  • System Prompt – Heavily guides the model to stick to BrainDrive-relevant responses.
  • Fine-Tuning – Focused on core facts that won’t change (e.g. our mission, how plugins work).
  • Context Injection (CAG) – Selectively adds docs into the model’s context window, based on what the user is asking about. Think of it like a smarter, pre-filtered retrieval system built into the backend.

:bullseye: Goals for V1

  • Works with small models (hopefully 1B but as small as possible if not) that run on local hardware.
  • Gives clear, helpful answers about BrainDrive.
  • Points people in the right direction when it hits a limit.
  • Easy to update as BrainDrive evolves.

:puzzle_piece: What’s Next

  • Finalizing the system prompt and training data.
  • Building the backend logic to manage context injection based on question type.
  • Designing the first interaction flow (what the user sees when they open BrainDrive for the first time).
  • Getting feedback from the community!

Matthew is going to be leading this project.

:light_bulb: How You Can Help

  • Writers/Docs People – Help refine the system prompt and core training text.
  • AI Builders – Share ideas or code for context window injection, local model selection, or fine-tuning.
  • UX Designers – Help us shape the first-run experience.
  • Testers – Once it’s ready, we’ll need help testing it on various devices.
  • Everyone – Got thoughts or ideas? Drop them in the thread below :backhand_index_pointing_down:

Let’s build a Concierge that doesn’t just work—but that shows off what it feels like to truly own your AI.

Here’s the recording: https://youtu.be/FnDBCAKlWzg

Thanks!
Dave W.

BrainDrive Concierge Development