Spaces:
Configuration error
title: README
emoji: 📊
colorFrom: purple
colorTo: blue
sdk: static
pinned: false
SmartTasks
Local-first desktop automation, powered by governance-validated AI.
SmartTasks builds tools that keep humans in command of their workflows — and of the AI those workflows increasingly rely on. Two things live here, tied together by the models we publish on this page.
🖥️ SmartTasks — Desktop Automation Platform
A no-code, local-first automation platform. Build powerful workflows with a visual drag-and-drop builder, connect any desktop app, and run everything natively on your own machine — full data privacy, no cloud upload required for execution.
- Visual drag-and-drop builder — triggers, actions, conditions, loops, sub-flows. No code.
- 14 core plugins + Marketplace — AI Tools, Data Storage, Scheduler, HTTP, Email, Webhook, and more.
- Local & secure execution — runs natively on Windows, macOS, and Linux; your data stays yours.
- AI-powered nodes — content generation, data analysis, and decision-making inside your flows.
- Advanced scheduling — cron, webhooks, file/app events, running 24/7.
🛡️ IAIso — The Governance Framework
As automation platforms hand more decisions to autonomous AI, IAIso provides the mechanical constraints that keep humans in command: hard-coded system invariants AI cannot cross, and human override built into the decision circuit. It's the governance layer behind everything we ship.
→ IAIso.org · github.com/SmartTasksOrg/IAISO
🤖 The Models on This Page
The GGUF models we publish here are the bridge between the two: local AI, validated against IAIso invariants, ready to power SmartTasks automation privately and on-device. Unlike a bare quantized re-upload, each ships with a documented, machine-readable governance profile so you can decide whether a model is fit for an autonomous loop before you wire it in:
- Capability scorecard — instruction-following, reasoning, structured output, and tool-calling, mapped to a capability tier (
scorecard.json, machine-readable). - Red-team security assessment — resistance to prompt-injection and jailbreak probes, reported honestly per attack class (weak spots included, not hidden).
- Transparency probe — tests for topic-suppression and viewpoint-alignment and reports findings verbatim. It discriminates — surfacing state-aligned framing in some models while clearing others as neutral.
- File integrity — SHA-256 for every quant, so you verify the exact weights you run.
| Model | Type | Notes |
|---|---|---|
| Qwen3-Coder-30B-A3B-Instruct-GGUF | MoE coder | Apache-2.0; agentic coding |
| Phi-mini-MoE-instruct-GGUF | MoE | MIT; clean transparency |
| Qwen3-4B-GGUF | Dense reasoning | Apache-2.0; L5 agentic |
More in the pipeline — text LLMs, MoE, and RAG components (embeddings + rerankers), each with the same governance treatment.
Local-first automation, human-in-command governance, and AI you can actually audit. Model cards are self-assessments documenting their own methodology and limitations.