Inside a 24/7 Autonomous Agent: Google AI, Tool Use, Langfuse & Obsidian

Chatbots wait for prompts; autonomous agents work continuously. In this Developer Showcase interactive booth, we peel back the layers of a 24/7 personal autonomous agent built on open-source Hermes Agent, powered by Google Gemini 3.7 Flash for fast reasoning and Google Nano Banana 2.1 for zero-shot visual generation for social media posts.

Rather than running canned, theoretical slides, attendees will interact with a live, persistent copilot handling daily real-world workflows—and explore how the architectural patterns powering this agent can be adapted for enterprise production agent deployments:

1. Tool Calling & Execution Engine: How the agent securely interfaces with local terminal environments, executes deterministic linting pipelines, and interacts with cloud SQLite/Turso vector databases.
2. Observability with Langfuse: Live tracing of agent decision loops, nested tool calls, latency breakdowns, and token economics to debug failures in real time.
3. Modular Skills & Memory Hierarchy: Dynamically extending agent capabilities with custom skill files, persistent cross-session memory, and deterministic guardrails.
4. Shared Obsidian Knowledge Hub: How human and agent co-author notes in a local Markdown vault, turning daily arXiv papers, RSS feeds, and travel research into an interconnected graph.
5. Real-World Live Workflows:
- Multi-Modal Content Creation: Autonomous technical research → structured post drafting → sketchnote generation via Google Nano Banana → visual verification.
- Autonomous Travel & Logistics: Itinerary research, destination databases, and personalized packing rules.
- Curated Reading & Bookmark Pipelines: Continuous vector-indexed ingestion of engineering blogs and book trackers.

Attendees will see live Langfuse trace waterfalls, inspect the agent’s inner loop in real time, and walk away with reusable architecture blueprints applicable to their own personal projects as well as production agent systems.