Configuration

Seepient merges configuration from four layers:

1. CLI command flags (highest precedence)
2. Workspace configuration file (.seepient/config.json in current directory)
3. User global configuration file (~/.seepient/config.json)
4. Environment variables
5. Built-in defaults (lowest precedence)

Managing settings interactively

The recommended way to configure providers and models is the interactive setup wizard:

seepient setup

The wizard writes validated configuration directly to ~/.seepient/config.json and manages credentials securely in your operating system keychain when available.

Configuration file format

Configuration files use standard JSON format. Below is an annotated example:

{
  "defaultProvider": "anthropic",
  "defaultModel": "claude-3-7-sonnet",
  "consentMode": "ask-untrusted",
  "sandbox": {
    "enabled": true,
    "allowNetwork": true
  },
  "routing": {
    "text": {
      "standard": { "provider": "anthropic", "model": "claude-3-7-sonnet" },
      "efficient": { "provider": "openai", "model": "gpt-4o-mini" },
      "complex": { "provider": "anthropic", "model": "claude-3-7-sonnet" }
    },
    "commit": {
      "standard": { "provider": "anthropic", "model": "claude-3-7-sonnet" }
    },
    "media": {
      "standard": { "provider": "fal", "model": "flux-pro" }
    }
  },
  "providers": {
    "anthropic": {
      "credentialRef": { "type": "env", "key": "ANTHROPIC_API_KEY" }
    },
    "openai": {
      "credentialRef": { "type": "env", "key": "OPENAI_API_KEY" }
    },
    "local": {
      "type": "openai-compatible",
      "baseUrl": "http://127.0.0.1:11434/v1",
      "apiKey": "ollama"
    }
  }
}

Environment variables

LLM providers

Variable	Provider	Purpose
ANTHROPIC_API_KEY	Anthropic	API key for Claude models
OPENAI_API_KEY	OpenAI	API key for GPT models
GEMINI_API_KEY	Google	API key for Gemini models
GLM_API_KEY	Zhipu GLM	API key for GLM models
DEEPSEEK_API_KEY	DeepSeek	API key for DeepSeek models
OPENROUTER_API_KEY	OpenRouter	Multi-model routing gateway
FAL_KEY	Fal.ai	Image generation and diffusion models

Tool integrations

Variable	Integration	Purpose
TAVILY_API_KEY	Tavily	Real-time web search tool (web_search)
SMTP_HOST	SMTP Email	Mail server host (send_email)
SMTP_PORT	SMTP Email	Mail server port
SMTP_USER	SMTP Email	Username for authentication
SMTP_PASS	SMTP Email	Password or app password
WEBHOOK_URL	Notifications	Target endpoint for send_notification

System and security settings

Variable	Default	Purpose
SEEPIENT_CONSENT_MODE	always-ask	Consent mode: always-ask, ask-untrusted, or autonomous-trusted
SEEPIENT_SANDBOX	true	Set to false to disable OS process sandboxing (macOS/Linux)
SEEPIENT_AUDIT_LOG	~/.seepient/audit.log	Path to the append-only 0600 audit log file
SEEPIENT_CONFIG_DIR	~/.seepient	Directory for sessions, credentials, and settings

Consent modes

Seepient provides three consent modes that determine when user confirmation is requested before executing tools:

always-ask (default)

Every tool call that causes external side effects requires confirmation. Read-only operations like reading a file run automatically, while shell executions and file writes prompt for confirmation.

ask-untrusted

Known, trusted tools run automatically if they match pre-configured allowlists. Unrecognized shell commands or mutations outside the working directory still prompt for confirmation.

autonomous-trusted

Actions within the configured working directory run without confirmation. Useful for continuous integration runners, Docker containers, and batch scripts where no interactive terminal is attached.