anunix-exec

name	anunix-exec
description	Execute a command on a running Anunix VM on Jekyll via the HTTP API. Use when you need to run ansh commands, create tensors, inspect objects, or interact with Anunix programmatically.
argument-hint	<command>
disable-model-invocation	false
allowed-tools	Bash(ssh *) Bash(curl *)
when_to_use	run anunix command, execute on vm, tensor operation, model operation, shell command on anunix

Execute Command on Anunix VM

Send a command to a running Anunix instance on Jekyll via the HTTP API. The QEMU VM runs on Jekyll (x86_64 test target).

Arguments

• $ARGUMENTS — the ansh command to execute (e.g., sysinfo, tensor create default:/w 4,4 int8)

Steps

1. Execute via HTTP API — routed through Jekyll directly:

   ssh jekyll "curl -s -X POST http://localhost:18080/api/v1/exec \
     -H 'Content-Type: application/json' \
     -d '{\"command\": \"$ARGUMENTS\"}'"
   

2. Parse the JSON response. Format: {"status": "ok", "output": "..."}

3. If the response is empty or connection refused, the VM may not be running. Suggest /anunix-deploy first.

Available Commands

Object Tools

• ls [ns:path] — list namespace entries
• cat <path> — read object payload
• write <ns:path> <content> — create State Object
• inspect <path> — full object inspection
• search <pattern> — search payloads

Tensor Operations (RFC-0013)

• tensor create <ns:path> <shape> <dtype> — create tensor (dtypes: int8, uint8, int32, float32)
• tensor fill <ns:path> <pattern> — fill with zeros/ones/range
• tensor info <ns:path> — show shape, dtype, size
• tensor stats <ns:path> — BRIN statistics (mean, variance, sparsity, min, max)
• tensor slice <ns:path> <start> <end> — row-slice
• tensor diff <path-a> <path-b> — element-wise delta
• tensor quantize <ns:path> <dtype> — dtype conversion
• tensor search <predicate> — find by BRIN (e.g., sparsity>0.5, dtype==int8)
• tensor matmul <a> <b> <out> — matrix multiply
• tensor relu <input> [output] — ReLU activation
• tensor transpose <input> [output] — transpose 2D

Model Namespace

• model info <name> — show manifest
• model layers <name> — list layer tensors
• model diff <a> <b> — compare models
• model import <ns:path> <name> — import safetensors

System

• sysinfo — CPU, memory, network, PCI
• mem stats — page allocator
• netinfo — network config
• help — full command list

Chaining Commands

for cmd in "tensor create default:/a 4,4 int8" "tensor fill default:/a ones" "tensor stats default:/a"; do
  ssh jekyll "curl -s -X POST http://localhost:18080/api/v1/exec \
    -H 'Content-Type: application/json' \
    -d '{\"command\": \"$cmd\"}'" | python3 -c "import sys,json; print(json.load(sys.stdin).get('output',''))"
done