Sidekick part 1: a terminal companion on your hardware
说明与用途
Sidekick 是一个本地终端助手,运行在本地硬件上,保持文件、内存和语音数据本地化,提供聊天、语音和 19 个工具,并支持审计和可选云集成。
问题
AI coding assistants typically require cloud accounts, upload code, and incur API costs, exposing local files to external servers.
结果
Sidekick provides a deterministic, grounded local AI assistant with full control over data, auditability, and optional cloud integration, eliminating the need for cloud accounts and reducing costs.
工作流
- Install Sidekick with "uv tool install sidekick-agent[voice]"
- Run "sk init" to guide first‑run hardware, model, and verification
- Start fullscreen chat with "sk"
- Use "sk talk" for voice input, transcribed locally
- Execute single‑shot tasks with "sk run "task"" and optional flags
- Access tools via "sk mcp" for stdio integration
- Review sessions, audit logs with "sk audit"
证据
Every AI coding assistant I used had the same deal: create an account, pipe your code through someone's cloud, watch the meter run.
Sidekick is the opposite bet: a local-first terminal companion you talk to — chat, voice, and 19 tools — running on your own hardware.
Your files, memory, and voice never leave your machine unless you hand it a key.
One command, sk, with several surfaces sharing everything: ... Fullscreen TUI, Plain REPL, Voice, Single-shot runs, MCP server.
Underneath: long-term memory with full-text search, todos, shell history, background daemon with calendar schedules, skill packs in SKILL.md format — and an audit ledger (sk audit) that logs every tool run, approval, and network egress so you can prove your code never leaked.
Bring your own key when you want frontier models — one guided sk connect flow, a dozen providers plus any OpenAI-compatible endpoint, spend caps per session, keys chmod 600 and masked.
- 记录日期: 2026-10-01T06:08:05.340Z