Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/yocxy2/RCLI/llms.txt

Use this file to discover all available pages before exploring further.

The rcli ask command sends a single text query to the LLM, executes any triggered actions, and prints the response. Perfect for scripting and automation.

​
Usage

rcli ask "<your query or command>"

# Examples
rcli ask "open Safari"
rcli ask "what is the capital of France?"
rcli ask "create a note called Ideas with buy milk"

​
How It Works

  1. Initialization — Loads LLM, STT, TTS models (one-time per invocation)
  2. LLM Processing — Classifies intent and generates response
  3. Action Execution — If action detected, executes via AppleScript/shell
  4. Response — Prints text to stdout
  5. TTS — Speaks response (unless --no-speak)
  6. Exit — Process terminates

​
Options

​
--models
string
default:"~/Library/RCLI/models"
Models directory path
​
--rag
string
Load RAG index for document-grounded answers
​
--gpu-layers
number
default:"99"
GPU layers for LLM
​
--no-speak
boolean
Disable TTS audio output
​
--verbose
boolean
Show debug logs

​
Examples by Category

​
Productivity Actions

rcli ask "create a note called Meeting Notes"
rcli ask "create a note called Ideas with buy milk, call mom"

​
Communication Actions

rcli ask "send message to John saying I'll be late"
rcli ask "text Sarah happy birthday"

​
Media Control

rcli ask "play some jazz on Spotify"
rcli ask "next track"
rcli ask "set music volume to 75"

​
System Control

rcli ask "open Safari"
rcli ask "quit Chrome"
rcli ask "list running apps"

​
Web & Navigation

rcli ask "search the web for RCLI documentation"
rcli ask "search YouTube for Python tutorial"

​
Conversational Queries

rcli ask "what is the capital of France?"
rcli ask "explain quantum computing in one sentence"
rcli ask "write a haiku about programming"

​
RAG-Powered Queries

Use --rag to query indexed documents: 
# Index documents first
rcli rag ingest ~/Documents/notes

# Query with RAG
rcli ask --rag ~/Library/RCLI/index "what were the key decisions?"
rcli ask --rag ~/Library/RCLI/index "summarize the project plan"
RCLI will:
  1. Embed the query
  2. Retrieve relevant chunks via hybrid search (vector + BM25)
  3. Pass context to LLM
  4. Generate document-grounded response

​
Scripting Examples

​
Shell Automation

#!/bin/bash
# morning-routine.sh

rcli ask "open Safari" --no-speak
rcli ask "open Mail" --no-speak
rcli ask "set volume to 30" --no-speak
rcli ask "what's on my calendar today?"

​
Capture Output

# Capture LLM response
response=$(rcli ask "what is 2 plus 2?" --no-speak 2>/dev/null)
echo "Answer: $response"
# Answer: Four.

​
Conditional Actions

#!/bin/bash
# Check battery, lock if low

battery=$(rcli ask "what's my battery level?" --no-speak 2>/dev/null)
if [[ $battery =~ ([0-9]+)% ]]; then
  level=${BASH_REMATCH[1]}
  if [ $level -lt 20 ]; then
    rcli ask "lock screen" --no-speak
  fi
fi

​
Performance

Typical latencies on Apple M3 Max:
  • Initialization — 1-2 seconds (model loading)
  • LLM TTFT — 15-25ms
  • LLM throughput — 150-200 tok/s
  • Action execution — 10-100ms (depends on action)
  • TTS synthesis — 100-200ms
For repeated queries, use interactive mode (rcli) to avoid reloading models.

​
Output Format

By default, rcli ask prints: 
Initializing...
<response text>
LLM: 87 tok  189 tok/s  TTFT 18ms
TTS: 142ms  0.8x RT
To get clean output (stdout only, no metrics): 
rcli ask "what is 2 plus 2?" --no-speak 2>/dev/null
# Four.

​
Action Execution Feedback

When actions are triggered: 
$ rcli ask "open Safari"
Initializing...
✓ open_app: Opened Safari
Done! Safari is now open.
LLM: 12 tok  203 tok/s  TTFT 15ms
TTS: 98ms  0.6x RT
The ✓ indicates successful execution. Failures show: 
✗ [failed] create_note  Error: Notes.app not installed

​
Error Handling

​
No Models Installed

$ rcli ask "hello"
Error: Models not found at ~/Library/RCLI/models
Run: rcli setup
Solution: rcli setup

​
Invalid Action Arguments

$ rcli ask "open"
I need to know which app to open. Try "open Safari" or "open Mail".
The LLM will ask for clarification.

​
Empty Query

$ rcli ask ""
Usage: rcli ask "your question or command"

​
Advanced Usage

​
Custom Context Size

# Increase context window for long queries
rcli ask "explain the entire history of computing" --ctx-size 8192

​
CPU-Only Mode

# Run LLM on CPU (slower, but frees GPU)
rcli ask "what is 2 plus 2?" --gpu-layers 0

​
Silent Mode (No TTS)

# Text output only
rcli ask "open Safari" --no-speak

​
Tool Calling Flow

When rcli ask "open Safari" is run:
  1. LLM receives: "open Safari"
  2. LLM generates: <tool_call>{"name": "open_app", "arguments": {"app_name": "Safari"}}</tool_call>
  3. Tool engine parses and executes open_app({"app_name": "Safari"})
  4. AppleScript runs: tell application "Safari" to activate
  5. Result: {"success": true, "output": "Opened Safari"}
  6. LLM generates final response: "Done! Safari is now open."
  7. TTS speaks: “Done! Safari is now open.”
Use --verbose to see full tool call JSON and execution logs.

Build docs developers (and LLMs) love

Get started for freeTalk to us