Chat & Slash Commands
Describe what you want in plain language. The agent decides which tools to use, does the work and reports back. Slash commands give you direct control.
Talking to the agent
Open a workspace and type in the chat box. Be specific about the result you want and where to put it:
Research the RTX 3060, 1660 Super and P100 GPUs. Compare them for LLM inference
and save your findings to "GPU-comparisons.md".There's a bug in "src/utils/parser.js": it crashes when the input is null. Find it and fix it.Read "meeting-notes.md" and create a todo for every action item.As the agent works, the conversation panel streams its reply live, along with each tool it calls and the result. Click Stop at any time to end the current turn.
Attaching files
Drag files into the chat box or use the attach button. Attachments are copied into the workspace's uploads/ folder so the agent can read them.
When the agent asks questions
For open-ended requests like "build me an app," the agent may ask a few questions before it starts. These questions come from
dialog templates in .agentos/DIALOGS.md, which you can edit:
## Dialog: write_content
Trigger: write a blog post, write an article, draft an article
Questions:
1. Who is the target audience?
2. What tone should the writing use?
3. Approximately how long should it be?You can also just say: "Before you start, ask me everything you need to know."
Slash commands
Type /help for the full list, or /help <command> for details. The most useful commands:
| Command | What it does |
|---|---|
/help [topic] | All commands, or details for one command, tool, skill or plugin |
/model · /model switch | Show or change the model |
/mode 1 · /mode 2 | Supervised or auto mode (see Jobs) |
/brain · /brain edit | View or edit BRAIN.md |
/memory show · /compact | Inspect or compact memory |
/jobs | List, approve, reject, run or halt jobs |
/cron | List, add or remove scheduled tasks |
/plan | Create and execute structured project plans |
/ralph-loop | Repeat a task until it's finished |
/install <repo> | Install skills, MCP servers, CLIs and tools from GitHub |
/credentials | Manage stored credentials (password-protected) |
/tools · /skills · /plugins | List what's loaded. Add reload to pick up changes from disk. |
/sessions | List sub-agent sessions |
/ls [path] · /cat <file> | List workspace files or print a file |
/prompt <file> | Send the contents of a file as your message |
/clear | Clear the conversation panel |
Sub-agents
For large or parallel work, the agent can spawn background sub-agents with their own context, then combine their results:
Research these five companies in parallel and save each report to its own file,
then write a comparison summary.Agents can also hand tasks to other workspaces' agents on the same machine, or to agents on the Agent Service Bus.
Tips
- Name the output file. "Save it to
research/output.md" gives you a lasting result. - Chain steps. "Read requirements.md, make a plan, then start on the highest-priority tasks."
- Switch to auto mode before long-running work so the agent doesn't wait for approvals.