Dashboard Tour

The DjinnBot dashboard at http://localhost:3000 gives you real-time visibility into everything your AI team is doing — from live agent output to cost analytics.

Home

The home page shows:

• System status — Redis connection, active runs, agent count
• Recent runs — latest pipeline and swarm executions with status
• Active sessions — live agent containers currently running
• Quick actions — start a new run, open chat, view projects, launch a swarm

Activity Feed

The activity page provides a live SSE-streamed feed of everything happening in the system:

• Agent wake/sleep events
• Task claims and transitions
• Pipeline step progress
• Tool call summaries
• Memory operations
• Quick stats (active sessions, tasks completed today, etc.)

The feed updates in real-time with no polling — events appear the instant they happen.

Runs

The runs page lists all pipeline and swarm executions. Click any run to see:

• Pipeline visualization — step-by-step progress through the pipeline
• Streaming output — real-time text output from the active agent
• Thinking blocks — expandable reasoning sections (for models that support it)
• Tool call groups — grouped tool invocations with arguments and results
• Step history — completed steps with full output and timing
• Key source badge — shows which API key was used (system, user, or agent override)

New Run Dialog

The new run dialog lets you:

• Select a pipeline or create a swarm
• Choose an agent and model
• Describe the task
• Link to a project
• Launch with one click

Swarm Runs

Swarm runs have their own dedicated visualization:

• DAG view — interactive directed acyclic graph showing task dependencies and parallel execution
• Status bar — overall progress, active/completed/failed task counts
• Timeline — chronological view of task execution with durations
• Task detail — click any task to see its output, agent, model, and timing

Chat

The chat interface lets you talk directly to any agent. Each chat session:

• Spawns an isolated container with the agent’s full toolbox
• Loads the agent’s persona and memories
• Supports multi-turn conversation with streaming responses
• Allows file uploads — drag and drop images, PDFs, or documents
• Supports image attachments sent as vision inputs to multimodal models
• Extracts text from uploaded PDFs automatically
• Shows key source badges indicating which API key is being used
• Supports inline HTML previews for agents that generate HTML output
• Groups related tool calls for cleaner display

Choose an agent and model from the header, then start typing. The floating chat widget provides quick access from any page.

Agents

The agents page shows all configured agents with:

• Status — current activity (idle, in pipeline, in chat, pulsing)
• Configuration — model, thinking level, pulse settings, coordination limits
• Pulse routines — view and manage named routines with independent schedules
• Built-in tools — enable/disable specific tools per agent
• Coordination — wake guardrails, concurrent session limits, messaging
• Recent runs — pipeline steps and pulse sessions this agent has executed
• Memory — browse and search the agent’s vault

Click an agent to edit their configuration — change models, enable pulse routines, adjust coordination settings, or modify tool access.

Projects

The projects page provides kanban-style project management:

• Board view — drag tasks between columns (Backlog, Ready, In Progress, Review, Done)
• Task details — description, acceptance criteria, assigned agent, linked PRs, review fields, work type classification
• Project vision — high-level project description and goals
• Planning pipeline — run the planning pipeline to auto-decompose projects into tasks (structured or agentic)
• Workflow policies — per-project SDLC stage routing rules that define which stages are required/optional/skip per task work type
• Code tab — interactive Sigma.js visualization of the project’s Code Knowledge Graph with community clustering, node details, and index management
• Pulse integration — agents autonomously pick up and work on Ready tasks during pulse cycles
• User provider settings — per-user API key configuration for cost control

Pipelines

Browse available pipeline definitions. View the YAML, see the step graph, and start new runs.

Skills

Manage agent skills — reusable instruction sets that agents can load on demand:

• Global skills — available to all agents (stored in agents/_skills/)
• Agent-specific skills — scoped to individual agents
• Skill packages — skills with reference files and HTML templates (e.g., visual-explainer)
• Enable/disable — toggle skills without deleting them
• Skill generator — AI-assisted skill creation via chat

MCP Tools

Configure MCP tool servers that agents can use:

• Server list — view all configured tool servers with status
• Health monitoring — live status (running, error, configuring)
• Tool discovery — see which tools each server provides
• Add servers — configure new MCP servers through the UI
• Hot reload — changes take effect immediately, no restart needed

Memory

Browse and search agent memory vaults:

• Personal vaults — each agent’s private memories
• Shared vault — team-wide knowledge
• Semantic search — find memories by meaning, not just keywords
• 2D graph view — interactive graph showing wiki-link connections
• 3D graph view — immersive WebGL visualization with Three.js — rotate, zoom, and explore memory clusters

Usage

The personal usage page shows per-user LLM API consumption:

• Token breakdown — input, output, and cache tokens per provider
• Cost tracking — estimated costs per model and provider
• Session history — which sessions consumed the most resources
• Key resolution — whether system or personal API keys were used

Admin Panel

The admin panel provides full operational control:

• API Usage — per-provider API call analytics with key source tracking (system vs. user keys)
• LLM Call Log — searchable log of every LLM API call with model, tokens, latency, cost, and approximate USD cost per call
• Container Logs — real-time streaming logs from all Docker containers
• Notifications — system notifications and alerts
• User Management — create, edit, and manage user accounts
• System Health — service status, active containers, and resource usage

Browser Cookies

Manage browser cookies for authenticated agent browsing via Camoufox:

• Upload cookies — upload Netscape-format cookie files (exported from your browser or the Cookie Bridge extension)
• Cookie sets — view, rename, and delete uploaded cookie sets
• Agent grants — grant or revoke cookie access per agent, controlling which agents can browse with your credentials
• Cookie Bridge extension — install the Chrome/Firefox extension to export cookies directly from your browser to DjinnBot

See Bot Interfaces for extension installation details.

Settings

Configure global and personal settings:

• LLM providers — add API keys for Anthropic, OpenAI, OpenRouter, and other providers
• Default models — set the default working model, thinking model, and Slack decision model
• Pulse settings — enable/disable autonomous pulse mode, set intervals
• Memory scoring — configure how memory relevance is calculated
• Secrets — manage encrypted secrets (GitHub tokens, SSH keys, etc.)
• User provider keys — configure personal API keys that override system keys
• Browser cookies — manage cookie sets and agent grants (also accessible from the dedicated Browser Cookies page)
• Two-Factor Authentication — enable/disable TOTP 2FA, view recovery codes
• API Keys — generate and manage API keys for CLI and programmatic access
• OIDC Providers — configure external identity providers for single sign-on