Configuration

PRX uses a TOML-based configuration system with hot-reload support. All settings live in a single file (with optional split fragments), and most changes take effect immediately without restarting the daemon.

Config File Location

The primary configuration file is:

~/.openprx/config.toml

PRX resolves the configuration directory in the following order:

1. OPENPRX_CONFIG_DIR environment variable (if set)
2. OPENPRX_WORKSPACE environment variable (if set)
3. Active workspace marker (~/.openprx/active_workspace.toml)
4. ~/.openprx/ (default)

The workspace directory (where memory, sessions, and data are stored) defaults to ~/.openprx/workspace/.

TOML Format

PRX configuration uses TOML -- a minimal, human-readable format. Here is a minimal working configuration:

# Provider and model selection
default_provider = "anthropic"
default_model = "anthropic/claude-sonnet-4-6"
default_temperature = 0.7

# API key (or use ANTHROPIC_API_KEY env var)
api_key = "sk-ant-..."

# Memory backend
[memory]
backend = "sqlite"
auto_save = true

# Gateway server
[gateway]
port = 16830
host = "127.0.0.1"

Configuration Sections

The configuration is organized into these top-level sections:

Section	Purpose
(top-level)	Default provider, model, temperature, API key
[gateway]	HTTP gateway: host, port, pairing, rate limits
[channels_config]	Messaging channels: Telegram, Discord, Slack, etc.
[channels_config.telegram]	Telegram bot configuration
[channels_config.discord]	Discord bot configuration
[memory]	Memory backend and embedding settings
[router]	Heuristic LLM router and Automix
[security]	Sandbox, resource limits, audit logging
[autonomy]	Autonomy levels and tool scope rules
[observability]	Metrics and tracing backend
[mcp]	Model Context Protocol server integration
[browser]	Browser automation tool settings
[web_search]	Web search and fetch tool settings
[xin]	Xin autonomous task engine
[reliability]	Retry and fallback provider chains
[cost]	Spending limits and model pricing
[cron]	Scheduled job definitions
[self_system]	Self-evolution engine controls
[proxy]	HTTP/HTTPS/SOCKS5 proxy settings
[secrets]	Encrypted credential store
[auth]	External credential import (Codex CLI, etc.)
[storage]	Persistent storage provider
[tunnel]	Public tunnel exposure
[nodes]	Remote node proxy configuration

See Configuration Reference for the full field-by-field documentation.

Split Configuration Files

For complex deployments, PRX supports splitting configuration into fragment files under a config.d/ directory next to config.toml:

~/.openprx/
  config.toml          # Main config (top-level + overrides)
  config.d/
    channels.toml      # [channels_config] section
    memory.toml        # [memory] and [storage] sections
    security.toml      # [security] and [autonomy] sections
    agents.toml        # [agents] and [sessions_spawn] sections
    identity.toml      # [identity] and [identity_bindings] sections
    network.toml       # [gateway], [tunnel], and [proxy] sections
    scheduler.toml     # [scheduler], [cron], and [heartbeat] sections

Fragment files are merged on top of config.toml (fragments take precedence). Files are loaded alphabetically.

How to Edit

Interactive Wizard

The onboarding wizard walks through provider selection, channel setup, and memory configuration:

prx onboard

CLI Config Commands

View and modify configuration from the command line:

# Show current configuration
prx config show

# Edit a specific value
prx config set default_provider anthropic
prx config set default_model "anthropic/claude-sonnet-4-6"

# Trigger a manual reload
prx config reload

Direct Editing

Open ~/.openprx/config.toml in any text editor. Changes are detected automatically by the file watcher and applied within 1 second (see Hot Reload).

Schema Export

Export the full configuration schema as JSON Schema for editor autocompletion and validation:

prx config schema

This outputs a JSON Schema document that can be used with VS Code, IntelliJ, or any editor that supports TOML schema validation.

Hot Reload

Most configuration changes are applied immediately without restarting PRX. The file watcher uses a 1-second debounce window and atomically swaps the live configuration on successful parse. If the new file has syntax errors, the previous configuration is retained and a warning is logged.

See Hot Reload for details on what requires a restart.

Next Steps

• Configuration Reference -- full field-by-field documentation
• Hot Reload -- what changes live vs. requires restart
• Environment Variables -- env vars, API keys, and .env support
• LLM Providers -- provider-specific configuration