Personal AI Infrastructure & Network
Autonomous local intelligence engine with tool-calling runtime, real-time web grounding, and persistent memory.
Caution
Use at your own risk. This software is developed primarily for personal workflows and experimental research. Certain implementations and architectural choices may be considered resource-intensive, expensive, or potentially unsafe outside a trusted personal environment:
- Host & Filesystem Access: The agent has tools to traverse, read, and OCR files across your local disk (
tools/files). It executes with the exact permissions of the user running the server. Never run PAIN as root. Be aware of indirect prompt injection risks from untrusted web pages or documents. - Heavy Compute & Thermals: Running local GGUF models with GPU offloading and up to 16k context budgets generates heavy sustained load on CPU, GPU, VRAM, and RAM. Ensure adequate cooling.
- Autonomous Tool Execution: The agent autonomously triggers web searches, evaluates math, queries files, and mutates persistent memory. Open-weights models can hallucinate or misinterpret prompts.
- Unencrypted Storage: Chat logs, prompt personalities, and persistent memory records are saved in unencrypted SQLite (
data/pain.db). - External Queries: While inference is strictly local, tools like
web_searchmake outbound internet queries.
Provided "as is" without warranty of any kind. The author assumes no liability for hardware damage, data loss, security exposure, or unintended actions.
- Local GGUF Runtime: Direct
llama-serverintegration with dynamic model switching. - Autonomous Tool Calling: Built-in web grounding, filesystem access, math, and OCR.
- Persistent Memory: SQLite FTS5 storage with live database inspection and CRUD.
- Real-Time Telemetry: Llama.cpp context gauge and one-click history compaction.
- Responsive UI: Minimal interface tuned for desktop, tablet, and mobile.
- Python 3.10+
llama.cppwithllama-servercompiled- GGUF models placed in
models/
# Clone and setup environment
git clone https://github.com/px7nn/PAIN.git
cd PAIN
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
# Configure paths in config.toml (point llama_dir to your llama.cpp build)
# Then start the server:
python3 main.py
Access the interface at http://localhost:8000.
PAIN's interface is fully responsive. To access your local model from your phone, tablet, or another machine:
- Local Network: Open
http://<LAN_IP>:8000from any device on your Wi-Fi/LAN. - SSH Tunnel:
ssh -N -L 8000:localhost:8000 user@remote-host - Tunnels & Mesh VPNs: Forward or route port
8000using Cloudflare Tunnel (cloudflared tunnel --url http://localhost:8000), Tailscale, or WireGuard to access your local agent from anywhere securely.
