Skip to main content

Agent orchestration platform — hardware-aware routing, MCP server, swarm intelligence, OpenAI-compatible

Project description

NeuralBroker 2.0 — Swarm-Native Orchestration Platform

NeuralBroker logo

VRAM-aware LLM routing · Zero-Trust Federation · MCP Server · Agentic Swarms

Python PyPI License: MIT FastAPI Website


The Ultimate Agent Orchestration Platform

NeuralBroker 2.0 transforms your machine into a Swarm-Native Orchestration Node. It goes far beyond standard model routing. It automatically evaluates your hardware, runs localized HNSW-vector memory, coordinates multi-agent task pipelines, and securely talks to other machines across the internet via Zero-Trust Federation.

pip install neuralbrok
neuralbrok start

That's it. Point Claude Code, Cursor, or any OpenAI-compatible app at localhost:8000/v1 (or connect via the built-in MCP server), and the Swarm takes over.


🌟 What's New in v2.0

1. Agentic Swarms & Task Coordination

NeuralBroker doesn't just route prompts; it routes objectives. The built-in SwarmCoordinator decomposes complex user requests into Plan → Execute → Review pipelines, automatically picking the best specialized local agent (Coder, Planner, Reviewer, Analyst) for each subtask.

2. NeuralFit Hardware Intelligence

Powered by a native Python implementation of our advanced hardware scoring algorithm, NeuralBroker scores models across Quality, Speed, Fit, and Context based on your exact hardware specifications (NVIDIA, Apple Silicon, AMD, or CPU). Run neuralbrok fit to see your live VRAM projections and tok/s estimates.

3. AgentDB (Vector Memory) & ReasoningBank

NeuralBroker now features a zero-dependency HNSW-style Vector Store built on NumPy. When your Swarms succeed at a complex task, the execution trajectory is saved into the ReasoningBank. Future agents automatically inject this historical context to solve similar problems faster.

4. Zero-Trust Federation (Agent-to-Agent Comms)

Swarms can now securely communicate across networks. Using our local mTLS/Ed25519-style crypto module, your agents can encrypt and sign payloads and send them to other NeuralBroker nodes. Untrusted nodes are automatically downgraded. Run neuralbrok federation init to start.

5. Security & AIDefence

Every prompt is now scrubbed.

  • PII Redaction: AWS keys, emails, and SSNs are automatically masked before reaching cloud LLMs.
  • Injection Shield: Federated incoming requests are heuristically scanned for adversarial jailbreaks ("ignore previous instructions").

6. Background Workers (Autopilot)

When your machine is idle, NeuralBroker wakes up. The internal WorkerDaemon launches background tasks to optimize codebases and run simulated security audits while you sleep.

7. MCP Server & Dynamic Plugins

NeuralBroker is now a native Model Context Protocol (MCP) server via stdio. Connect Claude Code directly to it. Want more agents? You can dynamically load .yaml agent definitions into your ~/.neuralbrok/plugins/ directory and share them with the community. Run neuralbrok plugins list to view.


🚀 Subscription Inheritance

You already pay for Claude Pro. NeuralBroker lets every app on your machine use it — automatically, for free. It reads the OAuth session your Claude Code CLI holds in ~/.claude/.credentials.json and uses the claude binary as a free inference backend for hard tasks that spill over from your local GPU.

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="any-string")
response = client.chat.completions.create(
    model="claude-sonnet-4-6",
    messages=[{"role": "user", "content": "Hello"}]
)
# Routed through your Claude Pro subscription. Cost: $0.

💻 Quickstart

1. Install & Start

pip install neuralbrok
neuralbrok start

2. Check Hardware Fit

neuralbrok fit

3. Start the MCP Server (For Claude Code/Cursor)

neuralbrok mcp

4. View Dashboard

Open http://localhost:8000/dashboard to view the live Pink Matrix telemetry UI, complete with routing waterfalls and VRAM gauges.


🔌 API & Integrations

NeuralBroker serves standard endpoints:

  • POST /v1/chat/completions (OpenAI format)
  • POST /nb/federation/receive (Zero-Trust node comms)
  • GET /nb/hardware (Telemetry)

Auto-Configured Integrations:

Run neuralbrok integrations setup <ide_name> to automatically connect:

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Cline
  • ...and 15+ more.

🛠️ Development

git clone https://github.com/khan-sha/neuralbroker.git
cd neuralbroker
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
pytest tests/

MIT License — see LICENSE.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

neuralbrok-2.0.6.tar.gz (150.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

neuralbrok-2.0.6-py3-none-any.whl (176.4 kB view details)

Uploaded Python 3

File details

Details for the file neuralbrok-2.0.6.tar.gz.

File metadata

  • Download URL: neuralbrok-2.0.6.tar.gz
  • Upload date:
  • Size: 150.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for neuralbrok-2.0.6.tar.gz
Algorithm Hash digest
SHA256 70d82c3a3db76cb07ceebb9b34f8428ba6cd85dba43b895249c32fca4a3683de
MD5 e23346a67857923cd11f3f8c34d173cb
BLAKE2b-256 6c11e63e3922d9777dc72a8a375baab296fa7cd6d11359537af5f3d2f8a8e78e

See more details on using hashes here.

Provenance

The following attestation bundles were made for neuralbrok-2.0.6.tar.gz:

Publisher: pypi-publish.yml on khan-sha/neuralbroker

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file neuralbrok-2.0.6-py3-none-any.whl.

File metadata

  • Download URL: neuralbrok-2.0.6-py3-none-any.whl
  • Upload date:
  • Size: 176.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for neuralbrok-2.0.6-py3-none-any.whl
Algorithm Hash digest
SHA256 bede225482700b984df68fafd6f47e32f17ec61d7550f056004e3700c3171d5f
MD5 64ace1b668a46c8e54fe9991e595c007
BLAKE2b-256 aa40860591b3829fd673f54edeec0f0bf885bb969db364c42df1769616f9f16d

See more details on using hashes here.

Provenance

The following attestation bundles were made for neuralbrok-2.0.6-py3-none-any.whl:

Publisher: pypi-publish.yml on khan-sha/neuralbroker

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page