Auggie is Augment Code's coding agent for the terminal. You describe work in natural language and it analyses the codebase, edits files, and automates routine tasks, using the same semantic codebase indexing that backs the company's IDE agents. Running auggie in a project directory opens an interactive session with streaming output and visible tool calls; auggie --print "your instruction" runs once and writes the result to stdout, and --quiet reduces that to the final output, so it can take piped input and be used in CI pipelines, serverless functions, and cron jobs. It needs Node.js 22 or later and a login.
Reusable prompts live as markdown files with frontmatter in a project's .augment/commands/ directory and become slash commands, for example /code-review path/to/file. For pull requests, the project publishes GitHub Actions that run the agent, review a PR, and describe a PR, added as workflows in a repository. Conversation history and task state persist across terminal sessions, work can be delegated to focused subagents that run in isolated contexts, and user and repository rules define guardrails that apply to every run.
Features
- Interactive session: real-time streaming, visible progress, and tool calls you can follow while the agent works
- Single-shot mode: --print runs one instruction and writes to stdout, --quiet returns only the final output, and input can be piped in
- Codebase context: the same semantic index used by the IDE agents rather than text search
- Custom slash commands: markdown files with frontmatter in .augment/commands/ become project-specific commands
- Subagents: specialized tasks such as security audits, test writing, or data analysis run in isolated contexts
- Session persistence: conversation history and task state carry across terminal sessions
- Rules: user and repository rules set guardrails the agent follows
- GitHub Actions: published actions run the agent, review a pull request, and describe a pull request from a workflow
