ask

Pipe-friendly AI for Unix and PowerShell pipelines. Sends stdin to a local Ollama model and writes the response to stdout — nothing else.

Install

npm install -g .

Requires Node 18+ and Ollama running locally.

Usage

ask [options] -p "prompt"

Option	Short	Default	Description
--prompt	-p	(required)	The instruction
--model	-m	gemma4:latest	Ollama model to use
--format	-f	text	Output hint: text json csv markdown
--system	-s		System prompt override
--temperature	-t	0.1	Sampling temperature
--raw			Skip output trimming and fence-stripping
--ollama-url		http://localhost:11434	Ollama endpoint

Config file

You can set defaults in ~/.ask-cli/config.json so you don't have to pass flags every time:

{
  "model": "llama3.2",
  "ollamaUrl": "http://192.168.1.10:11434"
}

Key	Description
model	Default model (overridden by -m)
ollamaUrl	Ollama endpoint (overridden by --ollama-url)

The file is optional — if it doesn't exist, the built-in defaults apply. CLI flags always take priority over config values.

Design principles

• stdout is sacred — only LLM output goes to stdout; status and errors go to stderr so pipes stay clean
• Streaming — text output starts printing as tokens arrive
• Deterministic — temperature defaults to 0.1 for consistent piped results
• Zero config — works out of the box if Ollama is running locally
• Stateless — each invocation is independent; context comes from stdin

Examples

# Review a diff before merging
git diff main | ask -p "review this diff for bugs, security issues, and unintended side effects"

# Generate a commit message from staged changes
git diff --staged | ask -p "write a concise git commit message for these changes"

# Chain models: find crash-looping pods, generate delete commands, run them
kubectl get pods -o json \
  | ask -p "which pods are in CrashLoopBackOff?" \
  | ask -p "generate kubectl delete commands for each" \
  | sh

# Mix with traditional tools
cat access.log | ask -p "extract all 5xx error lines" | grep -v "healthcheck"

# Generate summaries
cat README.md | ask -p "What does this program do?"

Format flag

The -f / --format flag tells the model what shape to produce. For json and csv, the full response is buffered and markdown fences are stripped automatically before writing to stdout, so downstream tools like jq and spreadsheet importers get clean input.

# Parsed by jq with no extra steps
echo "Paris, Tokyo, Lagos" | ask -p "as a JSON array" -f json | jq '.[0]'

# Ready to import as a spreadsheet
cat servers.txt | ask -p "convert to CSV with columns: hostname, role, os" -f csv > servers.csv

Model selection

Any model available in your local Ollama installation works:

ask -p "translate to French" -m gemma3:12b
ask -p "summarize" -m llama3.2
ask -p "extract entities" -m mistral

List available models:

ollama list