Configuration Reference

ravenbot is highly configurable via environment variables and a config.json file. This document describes all available settings.

🌍 Environment Variables (.env)

These are used for core system settings and credentials. See .env.example for the canonical reference.

AI Backend Settings

Variable	Description	Default
`AI_BACKEND`	`gemini` or `ollama`.	`gemini`
`GEMINI_API_KEY`	Required for Gemini. Get it from Google AI Studio.
`GEMINI_FLASH_MODEL` / `FLASH_MODEL`	Flash model for classification and simple tasks.	`gemini-2.5-flash`
`GEMINI_PRO_MODEL` / `PRO_MODEL`	Pro model for reasoning and complex research.	`gemini-2.5-pro`
`OLLAMA_BASE_URL`	API URL for your Ollama instance.	`http://localhost:11434/v1`
`OLLAMA_MODEL`	Default Ollama model for both tiers.
`OLLAMA_FLASH_MODEL`	Override Flash model for Ollama.
`OLLAMA_PRO_MODEL`	Override Pro model for Ollama.
`OLLAMA_THINKING`	Enable extended reasoning (if supported by model).	`false`
`OLLAMA_DISABLE_CLASSIFY`	Always use Flash model tier.	`false`

Integration Settings

Variable	Description
`TELEGRAM_BOT_TOKEN`	Your Telegram Bot API token.
`TELEGRAM_CHAT_ID`	Your specific Chat ID.
`DISCORD_BOT_TOKEN`	Your Discord Bot token.
`DISCORD_CHANNEL_ID`	Your specific Channel ID.
`JULES_API_KEY`	API Key for Jules Agent delegation.
`GITHUB_PERSONAL_ACCESS_TOKEN`	Required for GitHub MCP server features.
`BRAVE_API_KEY`	Required for web search via Brave Search MCP.
`VAULT_PATH`	Path to your personal Markdown knowledge base.

Web Server Settings

Variable	Description	Default
`WEB_PORT`	Port for the web UI.	`:8080`

Security Settings

Variable	Description	Default
`ALLOW_LOCAL_URLS`	Set to `true` to allow access to local/private IPs.	`false`

⚙️ config.json

The config.json file controls the internal logic, prompts, and scheduled jobs of the bot.

Bot Config (`"bot"`)

Field	Description
`systemPrompt`	Core persona. Routes to sub-agents and handles general conversation.
`researchSystemPrompt`	Specialized prompt for the ResearchAssistant sub-agent.
`systemManagerPrompt`	Specialized prompt for the SystemManager sub-agent.
`julesPrompt`	Specialized prompt for the Jules sub-agent.
`routingPrompt`	Prompt used to classify messages as Simple (Flash) or Complex (Pro).
`summaryPrompt`	Prompt used to compress conversation history into session summaries.
`flashTokenLimit`	Max tokens for the Flash model context window.
`proTokenLimit`	Max tokens for the Pro model context window.
`compressionThreshold`	Token percentage (0.0–1.0) to trigger async context summarization. Validated on load; defaults to 0.8 if out of range.

MCP Servers (`"mcpServers"`)

Defines the tools available to sub-agents. Each server is assigned to specific agents:

Server	Assigned To	Purpose
`brave-search`	ResearchAssistant	Web search
`weather`	ResearchAssistant	Weather lookups
`memory`	ResearchAssistant	Long-term memory
`filesystem`	ResearchAssistant	Server file access
`sequential-thinking`	ResearchAssistant	Step-by-step reasoning
`sysmetrics`	SystemManager	System metrics
`github`	Jules	GitHub API

Scheduled Jobs (`"jobs"`)

Defines automated research missions:

"jobs": [
  {
    "name": "Daily Briefing",
    "schedule": "0 0 7 * * *",
    "type": "research",
    "params": { "prompt": "Research the latest trends..." }
  }
]

🗄 Database

SQLite database stored at data/ravenbot.db (configurable via dbPath). Auto-migrated on startup. Key tables:

sessions — ADK session state
session_summaries — Compressed conversation summaries
reminders — Timed reminders with delivery tracking
watchers — Periodic monitoring tasks
daily_reports — Research report storage
vault_metadata — File checksum tracking
briefings — Daily briefing history
missions — Workflow mission tracking (pipeline, prompt, status, result, timestamps)