PIN Client Daemon (pin-clientd) v2.2.2

The PIN Client Daemon connects your local LLM inference server to the AiAssist P2P Inference Network.

Quick Start

1. Register as an operator at https://aiassist.net/pin/join
2. Copy the config template and add your credentials
3. Run the daemon: ./pin-clientd --config config.json

CLI Options

./pin-clientd [OPTIONS]

Options:
  -c, --config <FILE>     Config file path [default: config.json]
  -l, --log-level <LEVEL> Log level (trace, debug, info, warn, error) [default: info]
  -n, --threads <NUM>     Number of concurrent inference threads [default: 1]
  -h, --help              Print help
  -V, --version           Print version

Multi-threaded Inference

Use -n to enable parallel request processing:

# Process up to 4 requests concurrently
./pin-clientd -c config.json -n 4

Recommended values based on hardware:

• CPU-only: 1-2 threads
• Single GPU: 2-4 threads
• Multi-GPU: threads per GPU × number of GPUs

Important: Ollama requires additional configuration for parallel requests.

By default, Ollama processes requests sequentially. To enable parallel inference:

# Set before starting Ollama
export OLLAMA_NUM_PARALLEL=4
ollama serve

Or add to your systemd service file:

[Service]
Environment="OLLAMA_NUM_PARALLEL=4"

Backend	Parallel Support
Ollama	Requires OLLAMA_NUM_PARALLEL env var
vLLM	✅ Native (no config needed)
TGI	✅ Native (no config needed)
LMStudio	✅ Native (no config needed)

Match OLLAMA_NUM_PARALLEL to your daemon -n value for optimal throughput.

Installation

Prerequisites

• Rust (1.70+): Install via rustup

  curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
  source ~/.cargo/env

• LLM Backend (one of):

  • Ollama - Easiest for beginners
  • vLLM - Best for production GPU inference
  • text-generation-inference - HuggingFace's solution
  • LMStudio - Desktop app with API server

Build from Source

=======

Build in MSYS2:

rustup target add x86_64-pc-windows-gnu
rustup default stable-x86_64-pc-windows-gnu
cargo clean

Build

git clone https://github.com/aiassistsecure/pin-clientd.git
cd pin-clientd
./build.sh

// or simply
cargo build --release

Or manually with cargo:

cargo build --release
cp target/release/pin-clientd .

Setup

1. Register as Operator

Visit https://aiassist.net/pin/operator to register and get your credentials:

• clientId (starts with op_)
• apiSecret (starts with pin_sk_)

2. Create Config File

cp config.example.json config.json

Edit config.json with your credentials:

{
  "clientId": "op_your_id_here",
  "apiSecret": "pin_sk_your_secret_here",
  "nodes": [
    {
      "alias": "My-GPU",
      "inferenceUri": "http://localhost:11434",
      "apiMode": "ollama",
      "region": "us-east",
      "capacity": 5,
      "pricePerThousandTokens": 0.001
    }
  ]
}

3. Start Your LLM Backend

# For Ollama
ollama serve

# Verify it's running
curl http://localhost:11434/api/tags

4. Run the Daemon

./pin-clientd -c config.json

With multi-threading:

./pin-clientd -c config.json -n 4

5. Run as System Service (Linux)

Create /etc/systemd/system/pin-clientd.service:

[Unit]
Description=PIN Client Daemon
After=network.target ollama.service

[Service]
Type=simple
User=your-username
WorkingDirectory=/home/your-username/pin-clientd
ExecStart=/home/your-username/pin-clientd/pin-clientd -c config.json -n 4
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

Enable and start:

sudo systemctl daemon-reload
sudo systemctl enable pin-clientd
sudo systemctl start pin-clientd
sudo journalctl -u pin-clientd -f  # View logs

Configuration

Config Structure

{
  "clientId": "op_your_operator_id",
  "apiSecret": "your_api_secret",
  "nodes": [
    {
      "alias": "GPU-1",
      "inferenceUri": "http://localhost:11434",
      "apiMode": "ollama",
      "region": "us-east",
      "capacity": 10,
      "pricePerThousandTokens": 0.001
    }
  ]
}

Root Fields

Field	Required	Description
clientId	Yes	Your operator ID (starts with op_)
apiSecret	Yes	Your API secret from registration
nodes	Yes	Array of node configurations (at least one required)

Node Fields

Field	Required	Description
alias	Yes	Friendly name for this node
inferenceUri	Yes	LLM server URL (e.g., http://localhost:11434)
apiMode	Yes	API format: ollama or openai
region	Yes	Geographic region (see table below)
capacity	Yes	Max concurrent requests
pricePerThousandTokens	No	Your price per 1K tokens in USD (default: $0.001)

API Modes

Ollama Mode

Use "apiMode": "ollama" for standard Ollama installations.

• Model discovery: GET /api/tags
• Chat endpoint: POST /api/chat
• Default port: 11434

{
  "alias": "ollama-node",
  "inferenceUri": "http://localhost:11434",
  "apiMode": "ollama",
  "region": "us-east",
  "capacity": 5,
  "pricePerThousandTokens": 0.001
}

OpenAI Mode

Use "apiMode": "openai" for OpenAI-compatible APIs:

• vLLM

• text-generation-inference (TGI)

• LMStudio

• LocalAI

• Any OpenAI-compatible server

• Model discovery: GET /v1/models

• Chat endpoint: POST /v1/chat/completions

{
  "alias": "vllm-node",
  "inferenceUri": "http://localhost:8000",
  "apiMode": "openai",
  "region": "us-west",
  "capacity": 20,
  "pricePerThousandTokens": 0.002
}

Regions

Choose the region closest to your server's physical location.

Region ID	Name	Example Use Case
us-east	US East	AWS us-east-1, NYC, Virginia
us-west	US West	AWS us-west-2, California, Oregon
eu-west	EU West	AWS eu-west-1, Ireland, UK
eu-central	EU Central	AWS eu-central-1, Frankfurt, Amsterdam
asia-pacific	Asia Pacific	AWS ap-northeast-1, Tokyo, Singapore
global	Global (Any)	Multi-region or unknown location

Multi-Node Configuration

Register multiple nodes with different backends:

{
  "clientId": "op_abc123",
  "apiSecret": "secret_xyz",
  "nodes": [
    {
      "alias": "ollama-node",
      "inferenceUri": "http://localhost:11434",
      "apiMode": "ollama",
      "region": "us-east",
      "capacity": 10,
      "pricePerThousandTokens": 0.001
    },
    {
      "alias": "vllm-node",
      "inferenceUri": "http://localhost:8000",
      "apiMode": "openai",
      "region": "us-east",
      "capacity": 20,
      "pricePerThousandTokens": 0.002
    },
    {
      "alias": "lmstudio-node",
      "inferenceUri": "http://192.168.1.100:1234",
      "apiMode": "openai",
      "region": "us-west",
      "capacity": 5,
      "pricePerThousandTokens": 0.0005
    }
  ]
}

Running the Daemon

# Basic usage
./pin-clientd --config config.json

# With debug logging
RUST_LOG=debug ./pin-clientd --config config.json

# Specify log level
./pin-clientd --config config.json --log-level info

Interview System

When your daemon connects, the server sends interview prompts to verify LLM quality. The daemon automatically:

1. Receives test prompts from the server
2. Runs them against your local LLM
3. Reports timing metrics (TTFT, tokens/sec)
4. Gets assigned a quality tier

Quality tiers affect routing priority:

• verified - Highest priority (>90% accuracy, >20 tok/s)
• standard - Normal priority (>70% accuracy, >10 tok/s)
• slow - Budget tier (>70% accuracy, <10 tok/s)
• failed - Blocked from production (<70% accuracy)

Install as Service

sudo mkdir -p /opt/pin-clientd
sudo cp target/release/pin-clientd /opt/pin-clientd/
sudo cp config.json /opt/pin-clientd/
sudo cp pin-clientd.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable pin-clientd
sudo systemctl start pin-clientd

View Logs

journalctl -u pin-clientd -f

Example Configurations

File	Description
config.example.json	Basic single-node template
config.ollama.example.json	Ollama-specific example
config.openai.example.json	OpenAI-compatible API example
config.multi-node.example.json	Multi-node with mixed backends

License

MIT License - AiAssist Secure