Skip to content

configuration.md

Latest commit

 

History

History
258 lines (187 loc) · 7.67 KB

configuration.md

File metadata and controls

258 lines (187 loc) · 7.67 KB

Configuration

LLM providers, per-agent model overrides, failover, usage tracking, and all environment variables.

LLM Providers

Project Aether supports multiple LLM backends. Configure in your .env:

OpenRouter (Recommended — access to 100+ models)

LLM_PROVIDER=openrouter
LLM_API_KEY=sk-or-v1-your-key
LLM_MODEL=anthropic/claude-sonnet-4

OpenAI Direct

LLM_PROVIDER=openai
LLM_API_KEY=sk-your-openai-key
LLM_MODEL=gpt-4o

Google Gemini

LLM_PROVIDER=google
GOOGLE_API_KEY=your-google-key
LLM_MODEL=gemini-2.0-flash

Local Ollama (Free, private)

LLM_PROVIDER=ollama
LLM_MODEL=mistral:latest
# No API key needed

Other OpenAI-Compatible APIs

LLM_PROVIDER=together  # or groq, custom
LLM_API_KEY=your-key
LLM_BASE_URL=https://api.together.xyz/v1
LLM_MODEL=meta-llama/Llama-3-70b-chat-hf

Per-Agent Model Overrides

Optimize cost by using cheaper models for specific agents:

# Global (used by Architect)
LLM_MODEL=anthropic/claude-sonnet-4

# DS Team specialists use a cheaper model for script generation
DATA_SCIENTIST_MODEL=gpt-4o-mini
DATA_SCIENTIST_TEMPERATURE=0.3

Resolution order: UI model selection > per-agent .env setting > global default.

LLM Failover

Configure a fallback provider for resilience:

# Primary provider
LLM_PROVIDER=openrouter
LLM_API_KEY=sk-or-v1-your-key
LLM_MODEL=anthropic/claude-sonnet-4

# Fallback provider (used when primary fails after retries)
LLM_FALLBACK_PROVIDER=openai
LLM_FALLBACK_MODEL=gpt-4o

The circuit breaker opens after 5 consecutive failures and retries after a 60-second cooldown.

LLM Usage Tracking

Every LLM call is automatically tracked with token counts and estimated costs.

Dashboard

Navigate to LLM Usage in the sidebar to see:

  • Summary cards: total calls, tokens, estimated cost, models used
  • Daily trends: chart of token usage and cost over time
  • Per-model breakdown: usage and cost per model
  • Conversation costs: cost breakdown per conversation

API Endpoints

Method Endpoint Description
GET /api/v1/usage/summary?days=30 Usage summary with cost
GET /api/v1/usage/daily Daily usage breakdown
GET /api/v1/usage/models Per-model usage
GET /api/v1/usage/conversation/{id} Conversation cost

Custom Pricing

Pricing data covers OpenAI, Anthropic, Google, Meta, DeepSeek, and Mistral models. The pricing table is in src/llm/usage.py (previously src/llm_pricing.py).

Environment Variables

Complete reference of all environment variables from src/settings.py:

Core

Variable Default Description
ENVIRONMENT development Environment (development, staging, production, testing)
DEBUG false Enable debug mode

Database

Variable Default Description
DATABASE_URL — PostgreSQL connection string
DATABASE_POOL_SIZE 5 Connection pool size
DATABASE_MAX_OVERFLOW 10 Max overflow connections
DATABASE_POOL_TIMEOUT 30 Pool timeout (seconds)

Home Assistant

Variable Default Description
HA_URL — Home Assistant URL
HA_URL_REMOTE — Remote HA URL (for external access)
HA_TOKEN — Home Assistant long-lived access token

LLM

Variable Default Description
LLM_PROVIDER openrouter LLM provider (openrouter, openai, google, ollama, together, groq)
LLM_API_KEY — LLM API key
LLM_MODEL anthropic/claude-sonnet-4 Default LLM model
LLM_TEMPERATURE 0.7 Default temperature
LLM_BASE_URL — Custom base URL for OpenAI-compatible APIs
DATA_SCIENTIST_MODEL — Override model for DS Team specialists
DATA_SCIENTIST_TEMPERATURE — Override temperature for DS Team
LLM_FALLBACK_PROVIDER — Fallback LLM provider
LLM_FALLBACK_MODEL — Fallback LLM model
GOOGLE_API_KEY — Google Gemini API key

Observability

Variable Default Description
MLFLOW_TRACKING_URI http://localhost:5002 MLflow server URL
MLFLOW_EXPERIMENT_NAME aether MLflow experiment name

API

Variable Default Description
API_HOST 0.0.0.0 API bind address
API_PORT 8000 API port
API_WORKERS 1 Uvicorn worker count
PUBLIC_URL — Public URL for external access
API_KEY — API authentication key (empty = auth disabled in dev)

Authentication

Variable Default Description
AUTH_USERNAME admin Username for password login
AUTH_PASSWORD — Password for login (DB password from setup takes priority)
JWT_SECRET — JWT signing secret (auto-derived if empty)
JWT_EXPIRY_HOURS 72 JWT token expiry
GOOGLE_CLIENT_ID — Google OAuth client ID
GOOGLE_CLIENT_SECRET — Google OAuth client secret

CORS

Variable Default Description
ALLOWED_ORIGINS http://localhost:3000 Comma-separated allowed CORS origins

WebAuthn

Variable Default Description
WEBAUTHN_RP_ID localhost Relying party ID (your domain)
WEBAUTHN_RP_NAME Aether Display name
WEBAUTHN_ORIGIN http://localhost:3000 Full origin URL

Process Roles

Variable Default Description
AETHER_ROLE all Process role (all, api, scheduler)

Scheduler

Variable Default Description
SCHEDULER_ENABLED true Enable APScheduler
SCHEDULER_TIMEZONE UTC Scheduler timezone
WEBHOOK_SECRET — Secret for HA webhook validation

Trace Evaluation

Variable Default Description
TRACE_EVAL_ENABLED false Enable automatic trace evaluation
TRACE_EVAL_CRON 0 3 * * * Cron schedule for trace evaluation
TRACE_EVAL_MAX_TRACES 100 Max traces to evaluate per run

Discovery

Variable Default Description
DISCOVERY_SYNC_ENABLED false Enable periodic entity sync
DISCOVERY_SYNC_INTERVAL_MINUTES 60 Sync interval

Timeouts

Variable Default Description
TOOL_TIMEOUT_SECONDS 30 Default tool execution timeout
ANALYSIS_TOOL_TIMEOUT_SECONDS 120 Analysis tool timeout

Sandbox

Variable Default Description
SANDBOX_ENABLED true Enable gVisor sandbox
SANDBOX_TIMEOUT_SECONDS 30 Default sandbox timeout
SANDBOX_ARTIFACTS_ENABLED true Enable artifact collection from sandbox
SANDBOX_TIMEOUT_QUICK 15 Quick analysis timeout
SANDBOX_TIMEOUT_STANDARD 30 Standard analysis timeout
SANDBOX_TIMEOUT_DEEP 60 Deep analysis timeout
SANDBOX_MEMORY_QUICK 256m Quick analysis memory limit
SANDBOX_MEMORY_STANDARD 512m Standard analysis memory limit
SANDBOX_MEMORY_DEEP 1g Deep analysis memory limit

See Also