Proactive AI agent with long-term memory, dynamic LLM routing across 200+ models, and multi-channel delivery — Telegram, WhatsApp, CLI.
Semantic vector memory with intent-aware retrieval. Decides before each query if memory helps, rewrites for better recall.
200+ models, 5-tier cost scoring, bilingual EN+RU task detection. Picks the right model for each task automatically.
Cron jobs, heartbeats, daily digests, deadline reminders. Doesn't wait for you — reaches out when it matters.
From memory to channels to tools — a complete agent framework in a single Python package.
Automatically picks the best model for each task. Discovers 200+ models from OpenRouter, classifies by cost, and tracks success rates.
Single Python package. Async throughout. No microservices overhead.
┌────────────────────────────────────────────────────┐ │ Telegram │ WhatsApp │ Feishu │ CLI │ └──────┬─────────┴─────┬──────┴────┬─────┴─────┬─────┘ ▼ ▼ ▼ ▼ ┌────────────────────────────────────────────────────┐ │ Message Bus (async queues) │ └────────────────────────┬───────────────────────────┘ ▼ ┌────────────────────────────────────────────────────┐ │ Agent Loop │ │ Context · Tools (21) · Sessions · Skills │ │ Subagents · MCP Manager │ └──────┬─────────────┬─────────────────┬─────────────┘ ▼ ▼ ▼ ┌──────────────┐ ┌───────────┐ ┌─────────────────┐ │ LLM Gateway │ │ memU │ │ Cron + Heartbeat│ │ Router(5-tier)│ │ Memory │ │ Daily digest │ │ 200+ models │ │ Intent │ │ Reminders │ │ Fallback │ │ Proactive │ │ Scheduled tasks │ └──────────────┘ └───────────┘ └─────────────────┘
Three commands to a running agent.