CLI Usage

Complete reference for all command-line flags and options.

Table of Contents

  1. CLI Usage
    1. Synopsis
    2. Core Options
      1. –api-url, -a
      2. –model, -m
      3. –provider
    3. Prompts
      1. System Prompts
      2. User Prompts
    4. Configuration
      1. –config-file, -c
    5. Sampling Parameters
      1. –temperature, -t
      2. –max-tokens
      3. –seed
    6. Token Management
      1. –validate-tokens
      2. –context-limit
    7. Response Formatting
      1. –response-format
      2. –response-format-schema
      3. –response-format-schema-strict
    8. Authentication
      1. –api-key
      2. –api-key-name
    9. Network & Performance
      1. –timeout
    10. Output Options
      1. –output, -o
    11. Logging
      1. –verbose, -v
      2. –quiet, -q
    12. Input Validation (CLI-only)
      1. –enable-input-validation
      2. –max-input-length
      3. –max-input-tokens
    13. Complete Examples
      1. Example 1: Minimal Invocation
      2. Example 2: With All Common Options
      3. Example 3: OpenAI with JSON Schema
      4. Example 4: PDF Analysis with Guardrails
    14. Getting Help
    15. Next Steps

Synopsis 

1

fortified-llm-client [OPTIONS] --api-url <API_URL> --model <MODEL>

At minimum, you must provide:

  • --api-url (or via config file)
  • --model (or via config file)
  • One of: --user-text, --user-file, or --pdf-file

Core Options

–api-url, -a

Description: LLM API endpoint URL

Required: Yes (unless in config file)

Examples:

1
2
3
4
5
6
7
8

# Ollama (local)
--api-url http://localhost:11434/v1/chat/completions

# OpenAI
--api-url https://api.openai.com/v1/chat/completions

# Azure OpenAI
--api-url https://your-resource.openai.azure.com/openai/deployments/gpt-4/chat/completions?api-version=2024-02-15-preview

–model, -m

Description: Model name/identifier

Required: Yes (unless in config file)

Examples:

1
2
3
4
5
6
7
8

# Ollama
--model llama3
--model mistral
--model codellama

# OpenAI
--model gpt-4
--model gpt-3.5-turbo

–provider

Description: Force specific provider format (overrides auto-detection)

Values: openai, ollama

Default: Auto-detected from API URL

Example:

1

--provider openai  # Force OpenAI format even for Ollama-compatible URLs

Auto-detection analyzes the API URL to infer the provider. Explicitly set this only if auto-detection fails or you need to override it.

Prompts

System Prompts

Provide context or instructions to guide the LLM’s behavior.

–system-text

Direct text input:

1

--system-text "You are a Rust expert. Explain concepts clearly."

–system-file, -s

Read from file:

1

--system-file prompts/expert.txt

--system-text and --system-file are mutually exclusive. Use one or the other.

User Prompts

The main query or input to the LLM.

–user-text

Direct text input:

1

--user-text "Explain Rust ownership"

–user-file, -u

Read from file:

1

--user-file prompts/question.txt

–pdf-file, -p

Extract text from PDF (replaces user prompt):

1

--pdf-file document.pdf

--user-text, --user-file, and --pdf-file are mutually exclusive. Use only one.

Configuration

–config-file, -c

Description: Load default values from JSON or TOML file

Merge Priority: CLI arguments override config file values

Examples:

1
2
3
4
5
6

# TOML config
--config-file config.toml
-c config.toml

# JSON config
--config-file config.json

Config file example (config.toml):

1
2
3
4

api_url = "http://localhost:11434/v1/chat/completions"
model = "llama3"
temperature = 0.7
max_tokens = 2000

See Configuration for full config file documentation.

Sampling Parameters

–temperature, -t

Description: Sampling temperature (controls randomness)

Range: 0.0 to 2.0

Default: Provider default (usually 1.0)

Examples:

1
2
3

--temperature 0.0   # Deterministic (greedy sampling)
--temperature 0.7   # Balanced creativity
--temperature 1.5   # More creative/random

Lower = more deterministic, higher = more creative/random

–max-tokens

Description: Maximum response tokens to generate

Range: 1 to model’s max

Default: Provider default

Example:

1
2

--max-tokens 500   # Limit response to ~375 words
--max-tokens 2000  # Longer responses

–seed

Description: Random seed for reproducible sampling

Type: Unsigned 64-bit integer

Example:

1

--seed 42  # Same seed + temperature 0 = reproducible outputs

Token Management

–validate-tokens

Description: Enable token count validation before API call

Default: false

Example:

1

--validate-tokens

When enabled:

  1. Estimates tokens for system + user prompts
  2. Validates against context limit
  3. Fails early if exceeds limit (saves API cost)

See Token Management for details.

–context-limit

Description: Override model’s context window limit

Range: 1 to 9999999999

Default: Auto-detected from model name

Example:

1
2

--context-limit 8192   # Force 8K context limit
--context-limit 128000 # GPT-4 Turbo 128K context

Only needed if auto-detection fails or you want to enforce a lower limit.

Response Formatting

–response-format

Description: Control output format (OpenAI-compatible models only)

Values: text, json-object, json-schema

Default: text

Examples:

1
2
3
4
5
6
7
8

# Plain text (default)
--response-format text

# Unstructured JSON object
--response-format json-object

# JSON validated against schema
--response-format json-schema --response-format-schema schema.json

See Response Formats for complete guide.

–response-format-schema

Description: JSON Schema file path (required when --response-format=json-schema)

Example:

1
2

--response-format json-schema \
--response-format-schema schemas/product.json

Sample schema (schemas/product.json):

1
2
3
4
5
6
7
8

{
  "type": "object",
  "properties": {
    "name": {"type": "string"},
    "price": {"type": "number"}
  },
  "required": ["name", "price"]
}

–response-format-schema-strict

Description: Use strict mode for JSON schema validation

Default: true

Example:

1

--response-format-schema-strict false  # Disable strict mode

Authentication

–api-key

Description: API key for authentication (direct value)

Security Warning: Avoid using in shell history. Prefer --api-key-name.

Example:

1

--api-key sk-...

–api-key-name

Description: Environment variable name containing the API key

Recommended: Use this instead of --api-key

Examples:

1
2
3
4
5

# Set env var
export OPENAI_API_KEY=sk-...

# Use in CLI
--api-key-name OPENAI_API_KEY

--api-key and --api-key-name are mutually exclusive.

Network & Performance

–timeout

Description: Request timeout in seconds

Range: > 0

Default: 300 (5 minutes)

Examples:

1
2

--timeout 60    # 1 minute timeout
--timeout 600   # 10 minute timeout for large models

Output Options

–output, -o

Description: Write output to file instead of stdout

Features:

  • Atomic writes (temp file + rename)
  • Auto-creates parent directories
  • Overwrites existing files

Examples:

1
2

--output response.json
-o results/output.json

Logging

–verbose, -v

Description: Enable verbose logging (DEBUG level)

Conflicts with: --quiet

Example:

1
2

--verbose   # Show debug logs
-v          # Short form

–quiet, -q

Description: Suppress all logging output

Conflicts with: --verbose

Example:

1
2

--quiet  # No logs, only JSON output
-q       # Short form

Input Validation (CLI-only)

For LLM-based guardrails (Llama Guard, GPT OSS Safeguard, hybrid), use config files. These flags enable simple pattern-based validation only.

–enable-input-validation

Description: Enable regex-based input validation (PII, prompt injection, etc.)

Default: false

Example:

1

--enable-input-validation

When enabled, applies default pattern checks:

  • PII detection (SSN, credit cards, email addresses)
  • Prompt injection patterns
  • Suspicious characters/sequences

–max-input-length

Description: Maximum input length in bytes (requires --enable-input-validation)

Accepts: Human-readable sizes (100MB, 1.5GB, 500KB) or plain bytes

Default: 1048576 (1MB) when validation enabled

Examples:

1
2
3

--enable-input-validation --max-input-length 2MB
--enable-input-validation --max-input-length 500KB
--enable-input-validation --max-input-length 1048576  # Bytes

–max-input-tokens

Description: Maximum estimated input tokens (requires --enable-input-validation)

Default: 200000 (200K) when validation enabled

Example:

1

--enable-input-validation --max-input-tokens 100000

Complete Examples

Example 1: Minimal Invocation 

1
2
3
4

fortified-llm-client \
  --api-url http://localhost:11434/v1/chat/completions \
  --model llama3 \
  --user-text "Hello, world!"

Example 2: With All Common Options 

1
2
3
4
5
6
7
8
9
10
11

fortified-llm-client \
  --config-file config.toml \
  --api-url http://localhost:11434/v1/chat/completions \
  --model llama3 \
  --system-text "You are a helpful assistant." \
  --user-text "Explain Rust ownership" \
  --temperature 0.7 \
  --max-tokens 500 \
  --validate-tokens \
  --output response.json \
  --verbose

Example 3: OpenAI with JSON Schema 

1
2
3
4
5
6
7
8
9
10

export OPENAI_API_KEY=sk-...

fortified-llm-client \
  --api-url https://api.openai.com/v1/chat/completions \
  --model gpt-4 \
  --api-key-name OPENAI_API_KEY \
  --user-text "Generate a product catalog" \
  --response-format json-schema \
  --response-format-schema schemas/product.json \
  --output catalog.json

Example 4: PDF Analysis with Guardrails 

1
2
3
4
5
6
7
8
9

fortified-llm-client \
  --config-file guardrails-config.toml \
  --api-url https://api.openai.com/v1/chat/completions \
  --model gpt-4 \
  --pdf-file research-paper.pdf \
  --system-text "Summarize key findings" \
  --enable-input-validation \
  --max-input-length 5MB \
  --output summary.json

Getting Help 

1
2
3
4
5

# Show all options
fortified-llm-client --help

# Show version
fortified-llm-client --version

Next Steps