Changelog

All notable changes to the LLM Swarm project will be documented in this file.

[0.2.0] - 2026-06-02

Swarm API Key Authentication: Implemented shared-secret X-Swarm-Key header validation on all tracker and node endpoints. When SWARM_API_KEY is set, unauthorized requests are rejected with 401. Backward-compatible: unset key allows open access.
All outgoing node-to-tracker and node-to-node HTTP calls now include the X-Swarm-Key header automatically.
Updated docker-compose.yml to propagate SWARM_API_KEY to all services.

Initial Prototype: Core P2P Pooled Compute architecture implemented in Python.
Pipeline Parallelism: Logic for forwarding hidden state tensors between nodes.
Dynamic Peer Discovery: Centralized Tracker Node (tracker.py) for automatic node registration and health monitoring.
Docker Support: Added Dockerfile and docker-compose.yml for multi-platform deployment (M1 Mac, Linux, Windows).
Test Suite: Automated local mesh simulation script test_mesh.py.
Project Documentation: Comprehensive README with architecture overview and installation guides.
Model Slicing Utility: Added slice_model.py to allow nodes to host specific layer ranges of large GGUF models (e.g., Qwen-27B).
Agent Integration: Implemented OpenAI-compatible /v1/chat/completions endpoint for seamless use with Hermes Agent, AutoGPT, and other AI frameworks.
Public Swarm Instructions: Added detailed guides for users to join live swarms via public tracker URLs (e.g., ngrok).