Configuration

highvolt-server configuration is fetched from JSONAir at startup and refreshed periodically. The configuration is a single JSON document.

core

Key	Type	Default	Description
api_key	string	—	Internal API key (reserved)
mime_types.image	[]string	—	MIME types routed to the image LLM path
mime_types.pdf	[]string	—	MIME types routed to the PDF processor
mime_types.office	[]string	—	MIME types routed to the Office processor
mime_types.archive	[]string	—	MIME types routed to the archive processor
mime_types.text	[]string	—	MIME types routed to the text LLM path
minimum_image_size	int	5128	Minimum decoded image size (bytes); smaller images are skipped
max_pdf_pages	int	50	Maximum PDF pages to analyze per document
max_workers	int	2	Number of parallel worker goroutines
max_queue_size	int	50	Submission queue depth; excess submissions receive 503
queue_directory	string	—	Base directory for temp work files
temp_file_mode	string	"0600"	Octal file permissions for temp files
max_archive_size	int64	524288000	Max total extracted bytes from an archive (500 MB)
archive_extract_timeout	int	300	Max seconds for archive extraction (5 minutes)
max_body_size	int64	1073741824	Max HTTP request body size (1 GB)
export_command_timeout	int	120	Max seconds for PDF/Office conversion commands

http

Key	Type	Default	Description
listen	string	—	TCP listen address (e.g. :8443)
tls	bool	false	Enable TLS
cert	string	—	Path to TLS certificate (required if tls: true)
key	string	—	Path to TLS private key (required if tls: true)
mode	string	"production"	Gin mode: production, debug, or test. In production, HTTP request logging is disabled.

syslog

Key	Type	Default	Description
host	string	"local"	Syslog host, or "local" for stderr/stdout
proto	string	"tcp"	Transport protocol: tcp or udp

opensearch

Key	Type	Default	Description
url	string	—	OpenSearch base URL
username	string	—	Basic auth username
password	string	—	Basic auth password
index	string	—	Index name for documents
tls_skip_verify	bool	false	Disable TLS cert verification (not for production)

llm

Key	Type	Default	Description
url	string	—	LLM base URL (OpenAI-compatible). The path /chat/completions is appended automatically.
api_key	string	—	Bearer token for LLM API authentication
model	string	—	Model name (e.g. gpt-4o, llava:latest)
timeout	int	120	Request timeout in seconds
system_prompt	string	—	System prompt sent with every LLM request
user_prompt	string	—	User prompt sent with every LLM request

LLM prompt design

The LLM is expected to return a JSON object with this schema:

{
  "has_sensitive_data": true,
  "confidence": "high",
  "reasoning": "...",
  "description": "...",
  "status": "success"
}

Your system_prompt should instruct the model to always respond with valid JSON in this format. Example:

You are a privacy compliance assistant. You analyze documents for Personally Identifiable Information (PII). Always respond with a single JSON object containing: has_sensitive_data (boolean), confidence (low/medium/high), reasoning (string), description (string).

export_directories

Key	Type	Description
work	string	Base directory for temporary working files
archive	string	Directory for extracted archive content

Both directories must be writable by the process user.

export_commands

Key	Type	Description
pdf	string	Command to convert PDF to PNG images. Supports %INFILE%, %WORKDIR%, %OUTFILE%, %RANGE% placeholders.
office	string	Command to convert Office documents to PDF. Supports %INFILE%, %WORKDIR%, %OUTFILE% placeholders.

haproxy

Key	Type	Default	Description
enabled	bool	false	Enable HAProxy TCP health agent
port	int	—	TCP port for the HAProxy agent (required if enabled: true)