progresspals 0.1.2

Private AI clusters pooled over the internet, on your hardware. No VPS, no API fees.

progresspals

Private AI clusters pooled over the internet, on your hardware. No VPS, no API fees.

ProgressPals lets a small group of trusted people pool their GPUs and run massive open-source language models that no single machine could host on its own. One CLI, one invite link, one local OpenAI-compatible endpoint.

• 🌐 Website: https://progresspals.com
• 📚 Docs: https://progresspals.com/docs

Install

pip install progresspals

Requires Python 3.10+ on Linux or macOS. Windows is supported via WSL2.

Quick start

# 1. Authenticate (one-time)
pals init

# 2. Start a new private swarm
pals create Qwen/Qwen3-32B

# 3. Mint an invite for a teammate
pals invite create

# 4. They join with one command
pals join Qwen/Qwen3-32B <invite-token>

# 5. Expose an OpenAI-compatible endpoint at localhost:8080
pals serve

Then point Cursor, Aider, Continue, n8n, the OpenAI SDK, or anything that speaks /v1/chat/completions at http://localhost:8080/v1.

Supported models

Pass any of these HuggingFace model IDs directly to pals create:

Family	Sizes
Qwen 3	0.6B, 1.7B, 4B, 8B, 14B, 32B (dense), 30B-A3B, 235B-A22B (MoE)
Qwen 3-Coder	30B-A3B, 480B-A35B
Qwen 2.5	0.5B, 1.5B, 3B, 7B, 14B, 32B, 72B
Qwen 2.5-Coder	0.5B, 1.5B, 3B, 7B, 14B, 32B
Llama 2 / 3 / 3.1 / 3.3	up to 405B
Mixtral	8x7B, 8x22B
Falcon	40B, 180B
BLOOM	176B

How it works

The model is split across machines. Each peer holds a slice of the transformer layers. Inputs flow through the chain one peer at a time and stream back. Pipeline parallelism over libp2p, with per-swarm AES-256-GCM encryption on activation tensors between peers.

Swarms are invite-only. No public discovery, no leaderboard, no telemetry. Membership lives as a per-swarm allow-list and is enforced server-side every ~30 seconds.

What's inside

• Pipeline-parallel distributed inference for 8 model families
• libp2p peer discovery + NAT traversal
• Per-swarm AES-256-GCM activation encryption with HKDF-SHA256 key derivation
• Single-use, regenerable invite tokens
• Live peer dashboard (pals dash) — TUI with peer health, layer coverage, RPC stats
• OpenAI-compatible HTTP server (pals serve) — /v1/chat/completions, /v1/models, SSE streaming
• Member controls (pals peers list / kick, pals invite list / revoke)
• Optional INT8 / NF4 quantization via bitsandbytes for smaller VRAM footprints

License

Apache License 2.0. See LICENSE and NOTICE for full text and attributions.

Links

• Website: https://progresspals.com
• Documentation: https://progresspals.com/docs
• Contact: [e-mail]