Instar Persistent Agent Framework orchestrates long-running AI agents across multiple communication channels using Claude as the reasoning engine and Model Context Protocol tools for external integrations. The Agent Session Manager routes inbound events from Cron/Scheduler, Telegram, and WhatsApp to Claude Code Agent Runtime, which executes tasks via MCP tools for filesystem access, web search, and shell commands. Identity Store, Memory Store (Vector DB), Session Cache, and Persistent State DB maintain agent context and conversation history across sessions, while Logging/Observability tracks execution and File Storage persists artifacts. This architecture enables stateful, multi-turn agent interactions with deterministic tool access and audit trails—critical for production AI systems requiring reliability and traceability. Fork and customize this framework on Diagrams.so to adapt channel integrations, add new MCP tools, or modify persistence layers for your use case.