Your First Agent Task

This guide walks you through a complete agent interaction so you understand what's happening at each step.

1. Open the TUI

bun orginabox chat

The TUI connects to the opencode sandbox, creates a session, and subscribes to the SSE event stream.

The status bar shows:

⬢ orginabox chat  ·  session:a1b2c3d4  model:claude-sonnet-4-5  tokens:0  cost:$0.000

2. Send your first message

Type a message and press Enter:

> Summarise the top 3 news stories about AI this week and save the summary to memory.

The agent:

Uses browser_navigate to open a news source
Uses browser_snapshot to read the accessibility tree
Extracts headlines and summaries
Calls memory add fact "..." three times to persist results
Returns a formatted summary

3. Check what was saved

/memories AI

This queries the REST API at GET /v1/memories?q=AI and returns semantically-similar memories.

4. Start a new session using a template

/template list

Pick a template:

/template use research-analyst

Your next session will use the Research Analyst system prompt, which is optimised for deep web research, competitive intelligence, and structured reports.

5. Run a multi-task orchestration

/orchestrate Research the top 5 competitors of Salesforce, write a comparison table, and draft a 1-page executive summary.

The orchestrator:

Uses an LLM to decompose the task into subtasks
Runs independent subtasks in parallel sessions
Passes results from each subtask as context to dependent ones
Synthesises a final response

Check progress:

/plan-status <plan-id>

6. Review the audit log

/audit tool.execute

Every tool call, memory write, and auth event is logged. In the web dashboard, the admin panel shows a filterable audit log with full payloads.

Tips

Use /retry to resend the last message if a turn times out
Use /new to start a fresh session without losing the current one
Press Ctrl+K (web UI) or /help (TUI) to browse all commands
Memory is per-user; use /memories to inspect what the agent remembers about you