NovaBot is built as a practical operating foundation: focused agents, clear ownership, trusted communication, and technology chosen for work that needs to endure—not just impress in a demo.
Each layer has a job: agents coordinate the work, dependable services hold state and communication together, and shared operating practices make the whole environment portable and recoverable.
Hermes provides the operating surface. Individual bots maintain their own gateway, scope, memory, and responsibility—so each product can move with focus rather than a single general-purpose queue.
OpenAI and MiniMax provide complementary model access. The system can select the right capability for a task without making one provider a single point of failure.
Telegram and Hermes Desktop are the primary working surfaces. Dedicated Google accounts provide reliable service identity, while Resend handles outbound application email.
PostgreSQL supports shared, durable application data; SQLite keeps smaller and local-first systems simple. Mac mini and Ubuntu hosts are managed through established remote-desktop tools.
Every implementation belongs in a private GitHub repository and is pushed as changes are made. That creates a durable recovery path, a practical backup, and a clean way to migrate work to future tools or agents.
Market and portfolio systems draw on established brokerage, exchange, and data providers. Source choice is deliberate: higher-reliability feeds are preferred where rate limits make free sources unsuitable for repeatable operations.
An ops-only handoff added unnecessary friction. Product bots now own their approved deployments, with shared operations support available when it genuinely helps.
Discord proved too heavy for the client workflow. Telegram remains practical; Slack is the next communication platform to evaluate when the collaboration need justifies it.
New instances take real effort: API keys, messaging, identity, and tuning all need deliberate setup. Shared policy and reusable foundations reduce that startup cost.
yfinance is helpful for light exploration but too rate-limited for dependable recurring work. For production-grade usage, Finnhub is the preferred market-data path.
Private repositories, clear interfaces, and documented operating choices mean the work can move to a different AI provider, agent framework, or host without starting from zero.
Kanban remains available for work that benefits from shared visibility. It is not a required layer between a product owner and a tested, approved release.
