Local-first AI collaboration infrastructure. We build tools that let humans manage AI agents like a real team — with accountability, evidence chains, and human review at every critical step.
Every task delegated, every deliverable submitted, every review decision — recorded, auditable, on your own machine.
| Project | What it does |
|---|---|
| agent-mailbox | Local-first team workbench for AI agents: project mailboxes, shared context, task handoffs, human review. Think DingTalk for AI employees. |
| mcd-agent-gourmet | 麦当劳点餐 Skill — say "¥35, not spicy, no ice" → one answer + a copyable order card. Official pricing only. |
| dsh-devices | Turn your devices into a fleet — mDNS discovery, key pairing, SSH direct exec for dsh agents. |
| wx-fetch | WeChat article fetcher: Markdown / txt / JSON / Word output, zero dependencies. |
- Local-first — your data and evidence stay on your machine; no mandatory cloud.
- Evidence chains — who assigned, who accepted, what was delivered, who verified. Every step leaves a receipt.
- Human at the gate — agents work autonomously; humans review what matters.
- Ship, then polish — local versions lead public releases by 1–2 minor versions.
- Website: polaris-smart.github.io
- X (Twitter): @polaris_smart
- nofox.cn · Polaris-Smart LAB · 北极星智能实验室
Agent mailbox · AI employees · Human review · Local-first infrastructure