PSF Core closes the loop from natural-language intent to physical hardware execution. Where most orchestration tools stop at API calls, PSF Core drives PLCs, microcontrollers, and edge devices through a visual drag-and-drop pipeline that operators and integrators can run locally.
The Industrial Reflex Gateway (IRG) is where PSF Core goes further than anything else in the category. IRG bridges natural language intent to real hardware control, deterministically, safely, and at speed. A secure mod and plugin architecture with signed configurations ensures nothing runs on your infrastructure that you have not explicitly authorized.
Rounding out the platform: voice Speech-to-Text (STT) and Text-to-Speech (TTS) controls, coding and relay terminals, and access to the full Hugging Face model ecosystem, over 2.5 million models available for local deployment in just a few clicks. Open models install instantly with no API keys required. Gated and private models are fully supported too; bring your credentials and PSF Core handles the rest. No cloud dependency, no vendor lock-in. If it runs locally, PSF Core can run it.
PSF Core ships in Community Edition and Enterprise Edition, built on the same open foundation with local inference via llama.cpp and quantized GGUFs.
PSF Relay extends the PSF Core architecture into the physical world. By combining MoE routing, IRG execution, Bindings, and Channels, Relay translates natural language prompts into real device control, reliably and without abstraction theater.
Supported hardware spans the full range of practical deployment: microcontrollers including Raspberry Pi Pico/Pico 2, Arduino Uno, and ESP32; single-board computers including Raspberry Pi 2/3+/4/5 and Orange Pi 5+/6+; and integration patterns aligned with industrial PLC ecosystems from Allen-Bradley, Siemens, Schneider Electric, Omron, and Delta.
Start by flashing a tri-color LED on a Pico with a breadboard. Scale to driving an ESP32 robot truck wirelessly from a number pad. The architecture handles both without changing your mental model.
This repository hosts the PSF Core Community Edition codebase for community deployment, operations, and extension.
Instruction: Start here and choose the workflow card you need (Browse & Download, Catalog Editor, PSF Relay, PSF Terminal, or PSF Coding Terminal).
Instruction: Open a model card, review score/runtime metadata, then use Launch or Evaluate for quick validation.
Instruction: In Catalog Editor, click Add, paste HF model/page/download URLs, then use fetch actions to auto-fill metadata before saving.
Instruction: Use Binary Manager to check, download, repair, or remove runtime dependencies (Ollama, voice runtime, Node, llama.cpp, etc.).
Instruction: Open settings to configure HF key, speech/hardware, theme, system info, source control, and signed mod install/approval.
Instruction: Use terminal chat for model interaction; voice controls (Voice Off/PTT) and model dropdown allow fast runtime switching.
Instruction: Configure pipeline agents/channels/gateway/bindings, then deploy and run routed orchestration flows.
Instruction: Send IRG prompt contracts through full pipeline and inspect accepted contract + generated deployable code output.
Instruction: Keep Relay config visible while testing pipeline chat output to tune gateway/bindings and immediate execution behavior.
Instruction: Router/planner rewrites large requests into phased contracts before coder execution; review the plan before accepting output.
Instruction: Verify structured file block output (###FILE:*) to keep multi-file generation deterministic and directly runnable.
- Repository code: Apache-2.0 (see
LICENSE) - Commercial/private boundaries: see
EDITION_LICENSE_MATRIX.md
- Core-CE:
Core v<version> CORE-CE - Community Subscription:
Core v<version> COMMUNITY SUBSCRIPTION - Secure Community profile:
Core v<version> CORE-CE
Canonical naming and licensing policy is maintained in EDITION_LICENSE_MATRIX.md.
launcher/- desktop launcher/runtime and main platform modulesmodels/- catalog/config scripts and generated catalog JSON filesmods/- mod examples and mod tooling artifactsdocs/reference/- architecture, release, policy, and implementation docsinstall/- one-time bootstrap scripts for OS-specific prerequisitesrobot/- robot/relay firmware examples and pipeline profiles
- Community Edition: open-source local-first runtime for builders, labs, and field experimentation.
- Enterprise Edition: production support path for organizations scaling multi-team deployments and controlled workflows.
- Data Center Edition: hardened deployment profile with SELinux, Podman, and NGINX, plus pluggable inference backends such as vLLM, Triton/TensorRT-LLM, and TGI.
- Run the one-time OS bootstrap from the
install/directory. Linux/macOS:
cd install
./RUN_ONCE_MAC_LINUX.shWindows (PowerShell or CMD):
cd install
RUN_ONCE_WINDOWS.bat- Install launcher dependencies:
cd launcher
npm install- Run app (from workspace root):
./start.shWindows:
start.batmacOS:
./start.command- First-run checks:
- Open
Browse & Download Modelsand verify catalog loads. - Open
PSF Relayand load/deploy a pipeline profile if needed. - Open settings and confirm hardware/runtime detection is healthy.
main: integration branchfeat/*: active feature work- merge feature ->
mainwhen verified, then start next feature branch
- Preferred workflow: use the Settings gear (
API Keystab) to set/save your Hugging Face token - The app writes and reads the token from hidden
.envasHUGGINGFACE_TOKEN - The token is not persisted in
models/psf-settings.json .envis git-ignored in this repository
Reference .env format (advanced/manual use only):
# Local secrets only (git-ignored)
HUGGINGFACE_TOKEN=hf_your_real_token_here- Report vulnerabilities privately: see
SECURITY.md