Ethos: a TypeScript agent framework built around named specialist personalities

Ethos treats agent personality as a configurable, persistent unit rather than a one-off prompt. The TypeScript project, published as @ethosagent/cli, lets a developer select or scaffold a named personality, give it tools and memory, and then talk to it from the CLI or from messaging surfaces. The pitch is specialization over a single generalist agent. Five personalities ship by default, each with a defined working style: engineer — writes code and tests what ships researcher — seeks primary sources and flags uncertainty reviewer — separates blocking issues from suggestions coach — works through questions one at a time operator — dry-runs destructive actions, confirms, and documents outcomes Users can inspect a personality with ethos personality show, create new ones under ~/.ethos/personalities/, and keep durable notes in MEMORY.md and USER.md. Sessions, transcripts, and tool calls are browsable via ethos sessions, which matters for teams that need to audit what an agent did, not just what it said. The same identity is intended to follow across CLI, Telegram, Discord, Slack, and email. For multi-agent work, Ethos describes coordination through visible artifacts — a kanban board, a shared topic file, and an audit trail — instead of hiding decisions behind one conversational voice. That design choice is worth testing in your own workflow: it trades conversational smoothness for inspectability. Installation requires Node 24 or later on Linux, macOS, or WSL2. Native Windows is not a primary target. The standard flow is npm i