llamastash

name	llamastash
description	Control local llama.cpp-backed models through the LlamaStash CLI. Use when an agent needs to install or initialize LlamaStash, list models, start or stop a model, inspect daemon or proxy health, pull GGUFs from HuggingFace, or obtain the local OpenAI-compatible proxy address for harnesses like Claude Code, OpenClaw, OpenCode, Pi, and other shell-capable AgentSkills clients.
compatibility	Requires llamastash installed locally and available on PATH. Linux or macOS. Designed for Claude Code, OpenClaw, OpenCode, and similar shell-capable AgentSkills clients.
license	MIT
allowed-tools	Bash(llamastash *)
metadata	{"author":"deepu105","version":"0.1","openclaw":{"emoji":"🦙","requires":{"bins":["llamastash"]},"os":["darwin","linux"]}}

LlamaStash CLI

Use the llamastash CLI to install, inspect, and control local models managed by LlamaStash.

Current status

• Runtime status: !llamastash status --json 2>/dev/null || echo '{"error":{"code":"not_ready","message":"llamastash is not installed, the daemon is not running, or the CLI is not configured yet"}}'

First-time setup

If llamastash is missing or not configured yet, bring it up in this order:

1. Install the binary using the machine's native path:

   # macOS
   brew install llamastash/llamastash/llamastash
   
   # Linux
   curl -fsSL https://llamastash.dev/install.sh | sh
   
   # Portable fallback
   cargo install llamastash
   

2. Verify the binary:

   llamastash --version
   

3. Run the non-interactive first-run flow:

   llamastash init --recommended --json
   

4. Verify the install and branch on findings, not on doctor's exit code:

   llamastash doctor --json
   

When to use this skill

• The user wants an agent to manage local models through llamastash
• The agent needs the discovered model list before selecting a model
• The agent needs to start or stop a model and verify the result
• The agent needs daemon, host, GPU, or proxy state from status --json
• The agent needs to pull a GGUF from HuggingFace with llamastash pull
• The agent needs the local OpenAI-compatible base URL from proxy.listen
• The user wants LlamaStash installed or repaired on their machine

Key patterns

Prefer JSON everywhere it exists

For agent use, prefer:

llamastash init --recommended --json
llamastash doctor --json
llamastash list --json
llamastash status --json
llamastash recommend --json
llamastash pull <owner/repo[:filename.gguf]> --json

Do not parse the human-readable table or colored output.

Resolve exact model names first

Before start or stop, read the catalog first and reuse the exact discovered name from list --json.

llamastash list --json
llamastash start <exact-model-name>
llamastash status --json

start and stop are not the primary machine contract surfaces. Confirm the result with status --json.

Use doctor findings, not its exit code

llamastash doctor --json always exits 0. Escalate when findings is not empty.

Read the proxy address from CLI state

When wiring another harness to the built-in OpenAI-compatible proxy, read the bound address from status --json and build the base URL as http://<proxy.listen>/v1.

Default mode usually lands on 127.0.0.1:11435. Ollama-compat mode usually lands on 127.0.0.1:11434. If the base port is busy, LlamaStash may bind the next free port in the documented scan window, so check the live value first.

Branch on exit codes

Important codes:

Code	Meaning
64	bad CLI usage
65	daemon unreachable
66	model reference matched zero or multiple models
67	launch failed
68	stop failed
69	pull failed
70	llama-server binary not found
71	unexpected error
72	init aborted before substantive work
73	init download failed
74	init smoke failed

See references/commands.md for command patterns and examples. See INSTALL.md#for-ai-agents for agent installation patterns and examples.