Configuration Reference

This reference documents all configuration options for Praxis components.

Environment Variables

RabbitMQ

Database (Service)

Formats:

• postgresql://user:pass@host:5432/dbname - PostgreSQL
• sqlite:///path/to/file.db - SQLite with URL prefix
• /path/to/file.db - SQLite (implicit)

See Database Configuration for detailed setup.

Service

Build

Logging

Service Configuration

Service configuration is stored in the database and managed via the praxis TUI.

Application Logging

When disabled or missing, logging is off by default. The service broadcasts the current setting to nodes and clients at startup and on registration.

Log Query

LLM Provider Settings

Access via Settings (Ctrl+S) > LLM in the praxis TUI.

LLM configuration has two levels: a single list of named model definitions, and per-feature keys that each point at one definition by name.

Example llm_model_definitions value:

[
  {
    "name": "sonnet",
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514",
    "apiKey": "sk-ant-..."
  },
  {
    "name": "haiku",
    "provider": "anthropic",
    "model": "claude-haiku-4-5-20241022",
    "apiKey": "sk-ant-..."
  }
]

Each llm_feature_* key stores the name of one entry above (e.g. llm_feature_orchestrator = "sonnet"). A feature with no assigned model definition, or one whose name no longer resolves, is disabled.

Prompt Timeout

Claude Bridge Settings

Access via Settings (Ctrl+S) > Claude Bridge in the praxis TUI.

TLS is always on for both bridges; CCRv1 only accepts wss:// and CCRv2 only accepts https://. Leaf certs are minted per SNI on the fly and signed by a self-signed CA at ~/.praxis/bridge/ca_cert.pem.

The Claude Bridge allows Claude Code to connect directly to the service as a virtual node, without deploying a full Praxis node. See Claude Bridge for protocol details and setup instructions.

MCP Server Settings

Access via Settings (Ctrl+S) > Service in the praxis TUI.

The MCP server exposes all Praxis tools via the Model Context Protocol over streamable-HTTP transport. It is used by the built-in Orchestrator and can also be used by external AI agents. See MCP Server for full details.

Praxis Agent Settings

Access via Settings (Ctrl+S) > Agents in the praxis TUI.

Supported Providers

Every model definition can carry an optional base_url field that overrides the provider default. For custom the base URL is required — discovery and inference both fail without it. For ollama the base URL defaults to the local daemon; set it explicitly if you run Ollama remotely or on a non-default port.

Model Reference Format

When specifying models in operations or chains:

provider::model

Examples:

• anthropic::claude-sonnet-4-20250514
• openai::gpt-4o
• gemini::gemini-1.5-pro
• groq::llama-3.3-70b-versatile

Node Configuration

Node Commands

Nodes accept configuration commands at runtime:

Agent Connector Configuration

Each agent connector may have specific configuration. See individual connector documentation.

Claude Code

• Config path: ~/.claude/settings.json (global settings) and ~/.claude.json (preferences)
• MCP servers: ~/.claude/mcp.json, .mcp.json, and enabled plugin MCP definitions
• Plugins: ~/.claude/plugins/installed_plugins.json with cached components in ~/.claude/plugins/cache/
• Sessions: ~/.claude/projects/

Gemini CLI

• Config path: ~/.gemini/settings.json
• Sessions: ~/.gemini/tmp/<sha256-hash>/chats/

M365 Copilot

• Mode: DevTools (via CDP)
• Platform: Windows only

Claude Desktop

• Config path: %APPDATA%\Claude\claude_desktop_config.json (MCP servers), plus config.json and developer_settings.json in the same directory
• Sessions: none — Code/Chat are static UI modes driven via CDP, not session files
• Platform: Windows only

Codex CLI

• Config path: ~/.codex/config.toml (MCP servers), ~/.codex/auth.json (credentials)
• Sessions: ~/.codex/sessions/, ~/.codex/archived_sessions/

Cursor Agent

• Config path: ~/.cursor/cli-config.json (global), .cursor/cli.json / .cursor/mcp.json (project)
• Sessions: ~/.config/cursor/chats/<project_hash>/<chat_id>/ (SQLite store.db)

Droid CLI

• Config path: ~/.factory/settings.json, ~/.factory/mcp.json
• Sessions: ~/.factory/sessions/

Pi Coding Agent

• Config path: ~/.pi/agent/settings.json (no MCP support — extensions are the intended extension mechanism)
• Sessions: ~/.pi/agent/sessions/<encoded-cwd>/

Antigravity CLI

• Config path: ~/.gemini/antigravity-cli/settings.json
• Sessions: ~/.gemini/antigravity-cli/brain/

Operation Definitions

Operations are defined in JSON and stored in the service database.

JSON Format

{
  "item_type": "operation",
  "name": "find_credentials",
  "short_name": "find_credentials",
  "category": "recon",
  "description": "Search for hardcoded credentials",
  "agent_info": "Security researcher looking for exposed secrets",
  "timeout": 300,
  "operation_prompt": "Search the current directory for files that may contain hardcoded credentials, API keys, passwords, or secrets. List each finding with the file path and context.",
  "mode": "one-shot",
  "agent_iterations": 1,
  "yolo_mode": false,
  "disabled": false
}

Fields

Full Name

Operations are referenced by category::short_name, e.g., recon::find_credentials.

Chain Definitions

Chains are visual workflows stored in the service database.

Elements

block_config fields (all optional):

Session Groups

{
  "id": "group-1",
  "color": "#8B5CF6",
  "yolo_mode": true
}

Elements in the same session group share an agent session context.

Connections

{
  "id": "edge-1",
  "from_element": "trigger-1",
  "to_element": "op-1",
  "from_port": 0,
  "to_port": 0
}

condition is optional. Omit it entirely for a connection that always fires — when present, it must be OnSuccess or OnFailure. "Always" is not a valid value and fails to deserialize.

Intercept Rules

Rules for matching and processing intercepted traffic.

Rule Structure

{
  "name": "Capture API Keys",
  "regex_pattern": "Authorization:\\s*Bearer",
  "target_direction": "send",
  "scope": "all",
  "enabled": true,
  "summarization_prompt": "Extract and summarize the authentication tokens"
}

Target Direction

Scope

RuleScope carries no internal tag attribute (unlike sibling enums in the same file that do), so it serializes with serde's default external tagging rather than a {"type": "..."} shape.

Database Schema

SQLite (Default)

Default location: ~/.praxis/operations.db

Tables:

• service_config - Key-value configuration
• operation_definitions - Semantic operations
• operations - Operation executions
• operation_chains - Chain workflows
• chain_executions - Chain runs
• chain_triggers - Automated chain triggers (scheduled, intercept-match, new-node)
• chain_memories - Key-value store for chain Memory elements
• chain_payloads - Static content for Payload chain elements
• intercepted_traffic - Intercepted traffic
• intercept_rules - Traffic rules
• traffic_matches - Rule matches
• recon_results - Stored recon data
• event_log - Centralized logging table (controlled by application_logs_enabled)
• session_transactions - Per-prompt transaction records (request/response text, timing, status)
• lua_agent_scripts - Lua agent connector scripts (built-in and custom)
• toolkit_actions - Toolkit tool execution log
• remote_nodes - Persisted remote (virtual) node bridge configs
• agent_chat_sessions - AgentChat sessions
• agent_chat_agents - Agents participating in an AgentChat session
• agent_chat_channels - AgentChat channels
• agent_chat_messages - AgentChat channel and DM messages

PostgreSQL

For production and multi-instance deployments, use PostgreSQL. See Database Configuration for setup, migration, and tuning.

Default Ports

CLI Configuration

The Praxis CLI (praxis) stores state and can be configured via command-line options or its config file.

CLI State File

Contents:

{
  "client_id": "uuid-generated-on-first-run",
  "sessions": {
    "<node_id>": "<session_id>"
  }
}

sessions maps node IDs to the CLI's currently active ACP session ID on that node, populated by session create and consumed by session prompt / session close.

CLI Options

The RabbitMQ URL is never read from a -r/--rabbitmq flag or an OS environment variable — no such flag exists. It is read from ~/.config/praxis/config (key PRAXIS_RABBITMQ_URL). Use praxis set-rabbitmqurl <url> to persist it and praxis config to see the resolved URL and its source.

File Locations

Linux

macOS

Windows