CLI Commands

Complete reference for all predictor.sh CLI commands.

Global Flags

These flags work with all commands:

Flag	Description
--verbose, -v	Enable debug output
--version	Show version
--help, -h	Show help

---

Authentication

predictor login

Authenticate with predictor.sh.

predictor login [--token <key>]

Option	Description
--token <key>	Use API key directly (skip OAuth)

Without --token, opens browser for OAuth device flow.

predictor logout

Clear stored credentials.

predictor logout

predictor whoami

Show logged-in user.

predictor whoami

---

Configuration

predictor init

Generate predictor.yaml interactively.

predictor init

Prompts for model path, settings, and creates configuration file.

---

Tunnel Operations

predictor up

Start tunnel and serve model.

predictor up [options]

Argument	Description
<model>	Local path or HuggingFace model ID (auto-detected)

Option	Description	Default
--port <n>	Local port	8000
--cpu	Force CPU-only mode	false
--verbose	Enable debug output	false
--raw-logs	Output logs to stdout (no TUI)	false
--config <path>	Config file path	predictor.yaml

Examples:

# Local model
predictor up ./llama-7b.gguf

# HuggingFace model (auto-downloaded)
predictor up TheBloke/Llama-2-7B-GGUF

# Specific file from HuggingFace
predictor up TheBloke/Llama-2-7B-GGUF:llama-2-7b.Q4_K_M.gguf

# Force CPU
predictor up ./llama.gguf --cpu

# Debug mode
predictor up ./llama.gguf --raw-logs --verbose

predictor down

Stop tunnel gracefully.

predictor down [--force]

Option	Description
--force	Force-mark all endpoints as offline

predictor url

Print public endpoint URL.

predictor url

predictor status

Show endpoint status.

predictor status

---

Model Management

predictor pull

Download model from HuggingFace without starting server.

predictor pull <model-id> [options]

Argument	Description
<model-id>	HuggingFace model ID (required)

Option	Description
--file <pattern>	Download specific files
--revision <ref>	Specific revision/branch

Examples:

# Download default file
predictor pull TheBloke/Llama-2-7B-GGUF

# Specific file
predictor pull TheBloke/Llama-2-7B-GGUF --file "*.Q4_K_M.gguf"

# Specific revision
predictor pull TheBloke/Llama-2-7B-GGUF --revision main

predictor benchmark

Benchmark model performance.

predictor benchmark [options]

Argument	Description
<model>	Local path or HuggingFace model ID (auto-detected)

Option	Description	Default
--tokens <n>	Tokens to generate	128
--runs <n>	Benchmark runs to average	3
--json	Output as JSON	false

Examples:

# Basic benchmark
predictor benchmark ./llama.gguf

# More tokens, more runs
predictor benchmark ./llama.gguf --tokens 256 --runs 5

# JSON output for automation
predictor benchmark ./llama.gguf --json > results.json

---

Observability

predictor logs

View request logs.

predictor logs [options]

Option	Description	Default
--format <fmt>	Output format (human or json)	human
--follow, -f	Stream new logs	false
--since <duration>	Time range (e.g., "1h", "30m")	-
--limit, -n <n>	Max logs to show	-

Examples:

# Recent logs
predictor logs

# Follow mode
predictor logs --follow

# Last hour, JSON format
predictor logs --since 1h --format json

# Last 50 entries
predictor logs -n 50

predictor usage

Show current usage vs tier limits.

predictor usage

---

Token Management

predictor token show

Show endpoint auth token.

predictor token show [--reveal]

Option	Description
--reveal	Show full token (default: masked)

predictor token rotate

Generate new token (revokes old one).

predictor token rotate

This immediately invalidates the old token. Update all clients.

---

Exit Codes

Code	Meaning
0	Success
1	General error
2	Authentication error
3	Configuration error
4	Network error