CLI Reference¶

CRP includes a command-line interface for shell-based workflows and pipeline composition.

Installation¶

The CLI installs automatically with the SDK:

pip install crprotocol
crp --help

Commands¶

crp dispatch¶

Run a dispatch from the command line:

crp dispatch \
  --system "You are a technical writer." \
  --task "Write a guide to Kubernetes networking." \
  --provider ollama \
  --model llama3.1

crp init¶

Initialize a new CRP session with persistent CKF storage:

crp init --app-id my-project --ckf-path ./knowledge_base

crp status¶

Check session status:

crp status --session-id <session-id>

crp serve (HTTP Sidecar)¶

Start CRP as an HTTP service for polyglot deployments:

crp serve --port 9470 --provider ollama --model llama3.1

This exposes a REST API on http://localhost:9470 that any language can call. See HTTP Sidecar for details.

Pipeline Composition¶

CRP's CLI composes with standard Unix tools:

# Pipe document content through CRP
cat document.txt | crp dispatch \
  --system "Summarize this document." \
  --provider ollama --model llama3.1

# Process multiple files
for f in docs/*.txt; do
  crp dispatch --system "Extract key facts." --task "$(cat $f)" >> facts.txt
done

Environment Variables¶

Variable	Purpose	Example
`OPENAI_API_KEY`	OpenAI provider auth	`sk-...`
`ANTHROPIC_API_KEY`	Anthropic provider auth	`sk-ant-...`
`CRP_PROVIDER`	Default provider	`ollama`
`CRP_MODEL`	Default model	`llama3.1`
`CRP_CKF_PATH`	Default CKF storage path	`~/.crp/ckf`

Global Options¶

Flag	Description
`--verbose` / `-v`	Enable detailed logging
`--quiet` / `-q`	Suppress non-essential output
`--json`	Output in JSON format
`--version`	Show CRP version