ares-server 0.3.3

# A.R.E.S Configuration File
# ==========================
# This is the main configuration file for the Agentic Retrieval Enhanced Server.
# The server will NOT start without this file present.
#
# Configuration is hot-reloadable - changes to this file will be automatically
# detected and applied without restarting the server.

# =============================================================================
# Server Configuration
# =============================================================================
[server]
host = "127.0.0.1"
port = 3000
log_level = "info"

# =============================================================================
# Authentication Configuration
# =============================================================================
# Secrets are loaded from environment variables for security.
# Ensure these env vars are set before starting the server.
[auth]
jwt_secret_env = "JWT_SECRET"      # Env var containing JWT signing secret (min 32 chars)
jwt_access_expiry = 900            # Access token expiry in seconds (15 minutes)
jwt_refresh_expiry = 604800        # Refresh token expiry in seconds (7 days)
api_key_env = "API_KEY"            # Env var containing API key for service auth

# =============================================================================
# Database Configuration
# =============================================================================
[database]
url = "./data/ares.db"             # Local SQLite database path

# Optional: Turso cloud database (uncomment to enable)
# turso_url_env = "TURSO_URL"
# turso_token_env = "TURSO_AUTH_TOKEN"

# Optional: Qdrant vector database for semantic search
# [database.qdrant]
# url = "http://localhost:6334"
# api_key_env = "QDRANT_API_KEY"

# =============================================================================
# LLM Providers
# =============================================================================
# Define named provider configurations. These are referenced by models below.
# Only providers with compiled feature flags will be available at runtime.

[providers.ollama-local]
type = "ollama"
base_url = "http://localhost:11434"
default_model = "ministral-3:3b"

# NVIDIA DeepSeek v3.2 via OpenAI-compatible API (build.nvidia.com)
[providers.nvidia-deepseek]
type = "openai"
api_key_env = "NVIDIA_API_KEY"
api_base = "https://integrate.api.nvidia.com/v1"
default_model = "deepseek-ai/deepseek-v3.2"

# Example: Additional Ollama provider with different default model
# [providers.ollama-vision]
# type = "ollama"
# base_url = "http://localhost:11434"
# default_model = "qwen3-vl:2b"

# Example: OpenAI provider (requires 'openai' feature)
# [providers.openai]
# type = "openai"
# api_key_env = "OPENAI_API_KEY"
# api_base = "https://api.openai.com/v1"
# default_model = "gpt-4"

# Example: LlamaCpp for direct GGUF loading (requires 'llamacpp' feature)
# [providers.llamacpp]
# type = "llamacpp"
# model_path = "./models/granite4-tiny-h.gguf"
# n_ctx = 4096
# n_threads = 4
# max_tokens = 512

# =============================================================================
# Model Configurations
# =============================================================================
# Named model configurations that can be referenced by agents.
# Each model references a provider and specifies inference parameters.

[models.fast]
provider = "nvidia-deepseek"
model = "deepseek-ai/deepseek-v3.2"
temperature = 0.7
max_tokens = 256

[models.balanced]
provider = "nvidia-deepseek"
model = "deepseek-ai/deepseek-v3.2"
temperature = 0.7
max_tokens = 512

[models.powerful]
provider = "nvidia-deepseek"
model = "deepseek-ai/deepseek-v3.2"
temperature = 1.0
max_tokens = 8192

# DeepSeek model for complex reasoning tasks
[models.deepseek]
provider = "nvidia-deepseek"
model = "deepseek-ai/deepseek-v3.2"
temperature = 1.0
max_tokens = 8192

# Example: Vision model for multimodal tasks
# [models.vision]
# provider = "ollama-local"
# model = "qwen3-vl:2b"
# temperature = 0.7
# max_tokens = 512

# Example: Coding-specialized model
# [models.coding]
# provider = "ollama-local"
# model = "ministral-3:3b"
# temperature = 0.3
# max_tokens = 2048

# =============================================================================
# Tools Configuration
# =============================================================================
# Define which tools are available and their settings.
# Set enabled = false to disable a tool without removing it.

[tools.calculator]
enabled = true
description = "Performs basic arithmetic operations"
timeout_secs = 10

[tools.web_search]
enabled = true
description = "Search the web using DuckDuckGo (no API key required)"
timeout_secs = 30

# =============================================================================
# Agent Configurations
# =============================================================================
# Define agents with their models, tools, and system prompts.
# Agent types: router, orchestrator, product, invoice, sales, finance, hr

[agents.router]
model = "fast"
tools = []
max_tool_iterations = 1
system_prompt = """
You are a routing agent that classifies user queries and routes them to the appropriate specialized agent.

Available agents:
- product: Product information, recommendations, catalog queries
- invoice: Invoice processing, billing questions, payment status
- sales: Sales data, analytics, performance metrics
- finance: Financial reports, budgets, expense analysis
- hr: Human resources, employee information, policies
- orchestrator: Complex queries requiring multiple agents or research

Analyze the user's query and respond with ONLY the agent name (lowercase, one word).
Examples:
- "What products do we have?" → product
- "Show me last quarter's sales" → sales
- "What's our hiring policy?" → hr
- "Create a comprehensive market analysis" → orchestrator

Respond with ONLY the agent name, nothing else.
"""

[agents.orchestrator]
model = "powerful"
tools = ["calculator", "web_search"]
max_tool_iterations = 10
parallel_tools = false
system_prompt = """
You are an orchestrator agent that coordinates multiple specialized agents to answer complex queries.

Your capabilities:
- Break down complex requests into subtasks
- Delegate to specialized agents (product, invoice, sales, finance, hr)
- Synthesize results from multiple sources
- Perform web searches for external information
- Execute calculations when needed

Always provide comprehensive, well-structured answers.
"""

[agents.product]
model = "balanced"
tools = []
max_tool_iterations = 5
system_prompt = """
You are a Product Agent specialized in handling product-related queries.

Your capabilities:
- Product catalog search and recommendations
- Product specifications and details
- Inventory status and availability
- Product comparisons and alternatives
- Pricing information
- Product category navigation

Always provide accurate, helpful information about products.
If you don't have specific product data, suggest how the user can find it.
"""

[agents.invoice]
model = "balanced"
tools = ["calculator"]
max_tool_iterations = 5
system_prompt = """
You are an Invoice Agent specialized in invoice processing and billing queries.

Your capabilities:
- Invoice lookup and status
- Payment processing inquiries
- Billing history and statements
- Payment terms and due dates
- Invoice discrepancies and disputes
- Credit notes and adjustments

Always provide accurate billing information and calculations.
"""

[agents.sales]
model = "balanced"
tools = ["calculator"]
max_tool_iterations = 5
system_prompt = """
You are a Sales Agent specialized in sales data and analytics.

Your capabilities:
- Sales performance metrics
- Revenue analysis and forecasting
- Customer acquisition data
- Sales pipeline insights
- Regional and product-based breakdowns
- Commission calculations

Provide data-driven insights when possible.
"""

[agents.finance]
model = "balanced"
tools = ["calculator"]
max_tool_iterations = 5
system_prompt = """
You are a Finance Agent specialized in financial analysis and reporting.

Your capabilities:
- Financial statements and reports
- Budget tracking and variance analysis
- Expense management
- Cash flow analysis
- Financial projections
- Cost accounting

Ensure accuracy in all financial calculations and analyses.
"""

[agents.hr]
model = "balanced"
tools = []
max_tool_iterations = 5
system_prompt = """
You are an HR Agent specialized in human resources queries.

Your capabilities:
- Employee information and records
- Company policies and procedures
- Benefits and compensation
- Leave management
- Recruitment and hiring
- Training and development

Maintain confidentiality and provide policy-compliant responses.
"""

# =============================================================================
# Workflow Configurations
# =============================================================================
# Define how agents work together to handle requests.

[workflows.default]
entry_agent = "router"
fallback_agent = "orchestrator"
max_depth = 3
max_iterations = 5

[workflows.research]
entry_agent = "orchestrator"
max_depth = 3
max_iterations = 10
parallel_subagents = true

# =============================================================================
# RAG Configuration
# =============================================================================
[rag]
embedding_model = "BAAI/bge-small-en-v1.5"
chunk_size = 1000
chunk_overlap = 200

# =============================================================================
# Dynamic Configuration Paths (TOON Files)
# =============================================================================
# ARES uses a hybrid TOML + TOON configuration approach:
# - TOML (this file): Static infrastructure config (server, auth, database, providers)
# - TOON (config/*.toon): Dynamic behavioral config (agents, workflows, models, tools, MCPs)
#
# Benefits of TOON format:
# - 30-60% token savings when sending to LLMs
# - Hot-reloadable without server restart
# - One file per entity for easy management
# - LLM-friendly format optimized for AI consumption

[config]
agents_dir = "config/agents"       # Agent definitions (*.toon)
workflows_dir = "config/workflows" # Workflow definitions (*.toon)
models_dir = "config/models"       # Model configurations (*.toon)
tools_dir = "config/tools"         # Tool configurations (*.toon)
mcps_dir = "config/mcps"           # MCP server configurations (*.toon)
hot_reload = true                  # Watch for changes and reload
watch_interval_ms = 1000           # How often to check for changes