TARS
Local-first AI assistant
Tars is a compact Node.js assistant that runs on your machine under PM2. It connects a single active agent to Discord or a terminal, supports multiple model providers, and adds durable memory, explicit scheduled tasks, local skills, and MCP extensions.
It is aimed at operators who prefer a smaller bare-metal deployment and a clear local workspace. Projects such as OpenClawcover a broader gateway and channel ecosystem; Tars does not claim feature parity with multi-agent, mobile, voice, or broad multi-channel platforms.
Discord and terminal
Use one assistant from Discord or run a foreground terminal chat without starting duplicate daemon services.
Provider choice
Configure supported cloud providers or an OpenAI-compatible local endpoint through a validated configuration model.
Skills and MCP
Add trusted local instruction packages and explicitly enabled MCP servers. Extension subprocesses receive a restricted environment.
Durable memory and tasks
Keep selected facts and notes across restarts. Scheduled work runs only when a task is explicitly created; the heartbeat itself performs maintenance.
Defensive defaults
Discord owner checks, dashboard loopback binding, safer backup handling, outbound-fetch restrictions, and sensitive log redaction reduce common risks.
Operator-controlled
Tars runs with the permissions of its operating-system user. It is not a sandbox, so least-privilege accounts and review of tools and extensions remain essential.
$ npm install -g @saccolabs/tars
$ tars setup
$ tars start