nanobot

nanobot is a minimal, self-hosted AI agent designed to run wherever you work: a web UI, your terminal, or a chat interface. It's built for people who want a personal AI agent they control, without the overhead of bloated frameworks or cloud-locked platforms.

The core design philosophy is lean execution. Rather than consuming tokens without restraint, nanobot applies sensible context management and explicit token budgets, so your costs stay predictable. That matters especially for long-horizon tasks, where the agent needs to sustain coherent reasoning across tens or even hundreds of steps without losing the thread.

Key capabilities:

• Multi-surface operation: runs in a browser-based web UI, a terminal, or directly inside chat apps, so you're not locked into one interface
• Long-running workflows: maintains steady execution across complex, multi-step tasks without context drift
• Token budgeting: built-in controls keep spending predictable, unlike agents that leave cost management entirely to the user
• Embeddable runtime kernel: the agent core is portable, so you can drop it into a business workflow or a personal automation setup without rewiring everything around it
• Self-hosted: your data stays on your infrastructure; no third-party service holds your conversations or task history

Compared to heavier agent frameworks or fully managed platforms like Agenta, nanobot stays deliberately small. The codebase is compact, the runtime is portable, and the architecture doesn't assume you want a full orchestration platform. It suits developers and technically minded individuals who want an agent they can embed, extend, or just run locally without a lot of ceremony.

It supports automation tasks, developer integrations, and conversational use cases through the same core runtime. The MIT license means you can adapt it freely for personal or commercial use.