adk-dev-console

name	adk-dev-console
description	Explains the ADK Dev Console — what each tab shows, how to read Agent Steps, traces, multi-agent dashboard, agent switching, console modes, and other UI features visible at localhost:3001 during adk dev
license	MIT

ADK Dev Console

The Dev Console is a local app served at port 3001 (by default, but can be customized) during adk dev. It gives developers real-time visibility into their agent's behavior — conversations, execution traces, data, integrations, and configuration.

The Dev Console is a singleton shared dashboard: one UI server per user, with multiple adk dev agents registering and deregistering dynamically. Developers can switch between running agents in the sidebar, toggle between dev and prod targets, and connect to deployed Cloud bots — all from the same browser tab.

When to Use This Skill

Activate when users ask about:

• UI concepts — "What are Agent Steps?", "What does the Observe tab show?", "What is a turn?", "What is the component registry?", "What is the Agent Map?"
• Agent(0) UI — "What are the Tasks under Agent(0)?", "Where did the todo list come from?", "Why is Agent(0) showing task progress?"
• Dev Console navigation — "What tabs are available?", "Where do I find traces?", "How do I test RAG?"
• Reading execution data — "What do the steps mean?", "Why is my step red?", "What's the cost shown?"
• Component registry — "Where do I find custom components?", "How do I install a component?", "What components are available?"
• Multi-agent dashboard — "How do I switch between agents?", "How do I run multiple agents?", "What does the agent selector show?", "How do I see which agents are running?"
• Console modes and targets — "What's the difference between dev and prod?", "How do I connect to a Cloud bot?", "What is Cloud Dev Console?"
• Agent management CLI — "How do I list running agents?", "How do I stop an agent?", "What does adk ps show?", "How do I open the dashboard?"
• Specific pages — "How do I use the evals page?", "Where do I configure integrations?", "How do I search knowledge?"
• Comparing UI vs CLI — "Should I use the Dev Console or CLI for debugging?"
• Mentions of localhost:3001, "dev console", "Dev Console", "component registry", "agent selector", "agent picker", "agent map", "Cloud Dev Console", or specific tab names (Chat, Build, Components, Data, Test, Observe, Config)

Available Documentation

File	Contents
references/agent-steps.md	Agent Steps visualization — turns, iterations, tools, messages, state mutations, cost tracking, status indicators
references/pages.md	Every page/tab in the Dev Console — what it shows, key features, layout
references/component-registry.md	Component Registry — installed vs registry tabs, component lifecycle, runtime registry internals, UI features
references/multi-agent-dashboard.md	Multi-agent architecture, agent selector, console modes (local/cloud), dev/prod targets, CLI commands (adk ps, adk kill, adk dashboard, adk status)

How to Answer

Match depth to the question:

• "What is X?" (e.g., "What are Agent Steps?") → One sentence definition + what the user sees. Don't dump the full data model.
• "Where do I find X?" → Name the tab group and page, give the URL path.
• "How do I read X?" → Explain the visual hierarchy and what each element means.
• "What does this mean?" (pointing at something in the UI) → Identify the component, explain its meaning.

Quick Reference

Tab Groups

Group	Pages	Purpose
Chat	Chat	Test the agent via webchat + see Agent Steps
Build	Agent Map	Interactive agent architecture graph (experimental, feature-flagged behind enable_agent_forge)
Components	Webchat Components, Actions, Workflows, Triggers	Browse component registry, test bot primitives
Test	RAG Search, Evals	Test knowledge search and run automated evals
Data	Knowledge, Tables, Files	Manage knowledge bases, tables, and files
Observe	Conversations, Traces, Logs	View conversation history, execution traces, runtime logs
Config	Settings, Integrations	Agent config, secrets, LLM settings, integration management

Agent(0) Task Dock

When Agent(0) uses its todo tool, the panel shows a compact Tasks dock with completed/total count, the active task, and an expandable task list. Status icons map to pending, in progress, completed, and cancelled; priorities are shown as low/medium/high chips.

Multi-Agent Sidebar

The sidebar header shows the currently selected agent with a status indicator and mode pill (dev/prod/cloud). Clicking it opens the Agent Selector dropdown:

• Active agents — all running local agents with status dot, name, project path, and a close button
• Cloud Dev Console — switch to a deployed Botpress Cloud bot (workspace + bot picker)
• Recent projects — previously opened projects not currently running
• Footer actions — Create new project, Open existing, Switch environment (dev↔prod), About

Agent Steps (Chat Page)

The right panel of the Chat page shows Agent Steps — a real-time visualization of what the agent did to process each message.

Hierarchy: Turn → Iterations → Tools / Messages / State Mutations

• Turn = one conversation exchange (user message → agent processing → response)
• Iteration = one loop of the autonomous agent (think → decide → act)
• Tool = a tool call within an iteration (violet card)
• Message = a bot message sent (blue card)
• State Mutation = a state change (teal card, shows before/after)

Status indicators: ✓ green = ok, ✗ red = error, ⟳ blue spinning = running

AI metrics per iteration: model name, input/output tokens, cost (USD)

Key URLs

Path	Page
/chat	Chat + Agent Steps
/agent-map	Agent Map — interactive architecture graph (experimental)
/components	Component registry + installed components
/actions	Actions browser
/workflows	Workflows + run history
/search	RAG search testing
/evals	Eval definitions + runs
/knowledge	Knowledge base management
/tables	Table data management
/traces	Full trace viewer
/conversations	Conversation history
/logs	Runtime logs
/settings	Agent configuration
/integrations	Integration management

Multi-Agent CLI Commands

Command	Purpose
adk ps	List running agents and processes with PIDs, ports, uptime (supports --watch)
adk dashboard	Open the DevConsole in standalone mode (no agent required)
adk kill	Stop agents or the entire DevConsole (--all, --current, --pid, or by name)
adk status	Show project health and status info