Skip to content

SuperInstance/superinstance-ecosystem

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

36 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

SuperInstance Ecosystem

Teach once, run forever. The LLM never sees your shell.

Architecture

                              SuperInstance Stack
 ┌──────────────────────────────────────────────────────────────────┐
 │  User / Agent                                                    │
 │      │                                                           │
 │      ▼                                                           │
 │  ┌──────────────┐    ┌─────────────────┐    ┌────────────────┐  │
 │  │ lever-runner  │───▶│  Rust Carapace   │───▶│ tile-cuda      │  │
 │  │ (Python)      │    │  128ns hash      │    │  (CUDA + PTX)  │  │
 │  │ 202 tests     │    │  1.73µs embed    │    ├────────────────┤  │
 │  │               │    │                  │    │ tile-opencl    │  │
 │  └──────┬───────┘    └────────┬─────────┘    │  (OpenCL)      │  │
 │         │                     │               ├────────────────┤  │
 │         │                     │               │ tile-neon      │  │
 │         │                     │               │  (ARM NEON)    │  │
 │         │                     │               └────────────────┘  │
 │         │                     │                                   │
 │         │         ┌───────────┴────────────┐                     │
 │         │         │                        │                     │
 │         ▼         ▼                        ▼                     │
 │  ┌──────────────┐  ┌──────────────┐  ┌───────────────────┐      │
 │  │ lever-runner │  │ compiled-    │  │ fastloop-guard     │      │
 │  │ -wasm        │  │ policy-c     │  │ (Rust UDS cache    │      │
 │  │ (71KB gzip)  │  │ (µC / bare)  │  │  daemon)           │      │
 │  └──────────────┘  └──────────────┘  └───────────────────┘      │
 │                                                                  │
 │  ┌─────────────────────────────────────────────────────────────┐ │
 │  │ pincherOS — reflex caching, .nail state, 130+ tests        │ │
 │  ├─────────────────────────────────────────────────────────────┤ │
 │  │ tile-compiler — policy → GPU kernel pipeline, 109 tests    │ │
 │  ├─────────────────────────────────────────────────────────────┤ │
 │  │ open-minded — induction engine, tripartite synchronizer    │ │
 │  ├─────────────────────────────────────────────────────────────┤ │
 │  │ conservation-spectral-topology-rs — conservation laws (Rust)│ │
 │  ├─────────────────────────────────────────────────────────────┤ │
 │  │ captains-log — i2i coordination, session tracking          │ │
 │  └─────────────────────────────────────────────────────────────┘ │
 └──────────────────────────────────────────────────────────────────┘

 Hardware: Forgemaster (RTX 4050) + Loom (ARM64)

Repos

Repo What it does Lang Tests
lever-runner Injection-proof command runner, 70 tokens/cmd, skill packs Python 202
pincherOS Reflex caching, .nail state files, portable agent memory Python 130+
tile-compiler Policy → GPU kernel compilation pipeline Python 109
zeroclaw-arena Multi-agent arena for zero-trust competition
lever-runner-carapace Rust core: 128ns hash, 1.73µs embedding Rust
lever-runner-wasm Browser runtime, 71KB gzip WASM
tile-cuda CUDA + PTX kernel backend CUDA/PTX
tile-opencl OpenCL kernel backend (portable GPU) OpenCL
tile-neon ARM NEON SIMD backend (mobile/edge) C/ASM
compiled-policy-c Compiled policy for microcontrollers & bare metal C
ptx-bench PTX microbenchmarks, register pressure analysis PTX
fastloop-guard UDS cache daemon, hot-path acceleration Rust
conservation-spectral-topology-rs Conservation law verification, spectral topology Rust
open-minded Induction engine, tripartite synchronizer, code→knowledge 55+
intelligent-terminal Tripartite-classified terminal subsystem (MS fork) C++/C/Rust/JS
captains-log Cross-repo coordination, i2i session tracking
superinstance-ecosystem This repo — architecture, roadmap, integration

Real Numbers

Measured on real hardware. Not benchmarks. Not estimates.

Metric Value Where
Hash throughput 3.2M/sec (0.3µs) lever-runner-carapace
Embedding latency 1.73µs Rust Carapace
Vector search 14.6µs / 1K vectors lever-runner
GPU crossover 10K vectors (GPU wins) tile-cuda
WASM binary 71KB gzip lever-runner-wasm
Token budget ~70 / command lever-runner
Ecosystem tests 327+ across repos
Conservation violations 0 conservation-spectral-topology-rs

The Paradigm

  Traditional:  User → LLM → shell → chaos
  SuperInstance: User → lever-runner → verified command → result
                         ↑
                    LLM designs the command
                    but never touches the shell

The LLM proposes. lever-runner disposes. Every command is parameterized, bounded, auditable. You teach a skill once; it runs forever. No prompt injection, no escaping, no surprise rm -rf.

The GPU stack takes this further: policy compiled to CUDA/OpenCL/NEON kernels, a Rust carapace hitting 128ns hashes, and a UDS cache daemon for the hot path. The same agent code runs on an RTX 4050, in a browser via WASM, or on a microcontroller via compiled C.

Hardware

  • Forgemaster — x86 workstation, RTX 4050, CUDA host
  • Loom — ARM64, NEON target, edge inference

All repos · Architecture · Roadmap · Contributing

About

The agent operating system — four layers that compose

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors