Metadata-Version: 2.3
Name: nanowork
Version: 0.0.1
Summary: Tiny local-only agentic runner (placeholder, first release pending)
Author: Andrei Tim
Classifier: Development Status :: 1 - Planning
Requires-Python: >=3.12
Description-Content-Type: text/markdown

# NanoWork

> Placeholder release. The first real release is in preparation; this version contains no code.

**NanoWork** is a minimalistic orchestrator and runner of local atomic agents ("nanites") for non-coding work.
Local-first, solo-only, low-bloat, high-flexibility alternative to Claude Cowork, Codex, OpenWork, OpenWorker, etc.

## Principles

- **Simplicity and minimalism.** The smallest thing that works well. Nothing unused. Restraint-first.
- **No-chatter.** Concise structured output. Terse. Minimum words and mental overload. Just work done.
- **No-anthropomorphism.** No personality, no sycophancy. NanoWork is a tool. Scripting in English.
- **Cost minimisation.** Minimal context, optimal turns, high density. Cheap, fast, good for environment.
- **Visibility and clarity.** Every action, cost, and decision is surfaced with no clutter.
- **UX agnosticism.** The core isn't bound to any single I/O.

## Architectural Constraints

1. No LLM streaming. Why: simplicity, low chatter, structured I/O centric.
2. No cloud. No network except large LLMs, read-only crawls and search. Why: simplicity, safety.
3. Immutable atomic agents. No session edit, compact. Why: simplicity, small contexts centric.
4. SLMs. Local models with budget caps first, fine-tuned if needed. Why: cost, speed, safety.