Skip to content

PLUG_AND_PLAY.md

Latest commit

 

History

History
45 lines (32 loc) · 1.92 KB

PLUG_AND_PLAY.md

File metadata and controls

45 lines (32 loc) · 1.92 KB
Raw

PLUG_AND_PLAY — Pincher

A portable reflex runtime for agents. Teach, match, execute — learn as you go.

What Is This?

Pincher snaps into any shell and adds adaptive cognition. You teach it reflexes (intent → action pairs), it stores them as vectors in SQLite, and matches new intents against learned reflexes. It's a vector DB as runtime and LLM as compiler — no daemon, no cloud dependency.

Why Should You Care?

  • Works offline — ONNX embeddings run locally; hash-based fallback works without any model
  • Gets smarter with use — confidence scores update on every match; the system optimizes itself
  • Portable agent identity — pack your entire rig into a .nail file, move it to another machine
  • Sandboxed execution — bubblewrap isolation with veto engine prevents dangerous commands

Quick Start

git clone https://github.com/SuperInstance/pincher.git
cd pincher
cargo build --release -p pincher-cli
./target/release/pincher status

✨ Key Features

  • Teach → Match → Execute reflex engine with confidence scoring
  • SQLite-backed vector store (384-dim embeddings via sqlite-vec)
  • Bubblewrap sandbox with veto-based pre-execution blocking
  • .nail portable agent packing (tar.zst + BLAKE3 checksums)
  • CLI-driven: pincher do "...", pincher teach, pincher reflexes
  • JSON-RPC server for programmatic control

Next Steps

Guide What It Covers
GETTING_STARTED.md Build, run, teach your first reflex
ARCHITECTURE.md How it works under the hood
API_REFERENCE.md Every public function, type, and trait
LOW_LEVEL.md Internals, performance, porting guide

Status

v0.1.0 — Active development. Core reflex engine is production-grade. WASM guest execution is wired but experimental. Landlock sandboxing needs production testing.