OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Developer Tools & SDKs

Brainscope

Brainscope is a Python developer tool that serves Hugging Face language models through an OpenAI-compatible chat endpoint and visualizes model activations, attention and steering in a browser.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Brainscope — OpenViST project profile in AI Developer Tools & SDKs.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Brainscope when developing with a Hugging Face model and you want an OpenAI-compatible chat endpoint alongside browser-based inspection or steering of model activity.

Who it is for

  • Developers building applications with Hugging Face language models
  • Practitioners exploring model activations, attention or steering during chat generation

Documented uses

  • Route an OpenAI-client chat application to a Hugging Face model served locally.
  • Inspect logit, tuned-lens and J-lens readouts and attention during generation.
  • Apply per-request steering and examine how it affects model behavior.
  • Replay and analyze saved generations token by token.
Before you choose

Requirements & limits

The project documents pip installation and CPU and CUDA Docker images. Python installation requires Python 3.11 or later. The server has no authentication.

  • The interface is compatible with OpenAI-style chat clients while serving Hugging Face models.
  • The README documents CPU use with its tiny preset, as well as CUDA container instructions for larger models.
  • The README describes the middle-layer raw logit-lens view as approximate.
  • The server has no authentication, so the documented local port binding is relevant when running it.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Start the CPU-friendly demo server

Install Brainscope and launch its tiny-model server with the visualization interface.

Before you start

  • Python 3.11 or later
  • Internet access to install the package and retrieve model files

Steps

  1. Install the package with the command below.

    Documented command
    pip install brainscope

  2. Run brainscope --model tiny --lens on.

  3. Open the browser page launched by the server to use its built-in chat and view model activity.

Expected outcome

A local chat interface and browser visualization for the tiny, CPU-friendly model.

Guide limitations

  • The README describes the raw logit-lens view of middle layers as approximate.
  • The server has no authentication; the README advises keeping its port bound to the local machine unless using a trusted network.
Documentation 1 ↗
Source-based practical guide

Connect an OpenAI client to the local server

Route an existing OpenAI client to a running Brainscope server.

Before you start

  • A running Brainscope server
  • An application using the OpenAI client

Steps

  1. Set the client’s base_url to http://localhost:8010/v1.

  2. Send chat requests through that client.

  3. Open the Brainscope visualization page to inspect per-token and per-layer activity.

Expected outcome

The application sends chat requests through Brainscope, where supported model activity can be viewed in the browser.

Guide limitations

  • This recipe assumes the server uses its documented default port.
  • The visualization is an instrument for observing and steering model internals, not a claim of scientific findings.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
moudrkat/brainscope · Official homepage title · retained in its source language
How we review sourcesSuggest a correction