Your First Run

Now that DjinnBot is running, let’s complete initial setup and get agents working.

Initial Account Setup

If you enabled authentication (AUTH_ENABLED=true), the first time you open the dashboard you’ll be redirected to the setup page:

After completing setup, you’re logged in and ready to go. Additional users can be created through the admin panel or API.

If authentication is disabled, you skip this step and go straight to the dashboard.

The Three Workflows

DjinnBot has three complementary ways of getting work done:

Most of the time you’ll use projects. Pipelines power planning, structured output, and predefined workflows. Swarms are for when you want to throw multiple agents at a decomposed task set simultaneously.

Create a Project

Plan the Project

Once your project exists, decompose it into tasks:

1. Open your project and click Plan Project
2. This runs the two-stage planning pipeline:
   • Stage 1: Eric (Product Owner) breaks the project into high-level tasks with priorities and dependencies. Finn (Architect) validates the breakdown.
   • Stage 2: Eric decomposes into bite-sized subtasks (1-4 hours each). Finn validates the subtasks.
3. Tasks are automatically imported onto the kanban board with priority labels (P0-P3), dependency chains, hour estimates, and tags

The board starts with columns: Backlog, Ready, In Progress, Review, Done.

Assign Agents and Enable Pulse

Each agent is configured to watch specific board columns matching their role:

• Yukihiro (SWE) watches Ready — picks up implementation tasks
• Finn (Architect) watches Review — reviews implementations
• Chieko (QA) watches Review — tests implementations
• Stas (SRE) watches deployment-related tasks

Enable pulse mode for your agents via the dashboard (Agents page) or in each agent’s config.yml:

pulse_enabled: true
pulse_interval_minutes: 30
pulse_columns:
  - Ready

When pulse fires, agents autonomously:

1. Check the board for tasks in their columns
2. Claim the highest-priority task
3. Create a feature branch and spin up an isolated container
4. Do the work — write code, run tests, use tools
5. Open a pull request
6. Move the task to the next column

Run a Swarm

For maximum throughput on a set of independent tasks:

1. Click New Run in the dashboard
2. Select Swarm mode
3. Describe the overall objective
4. The planning agent decomposes it into a task DAG
5. The swarm executor runs multiple agents in parallel, respecting dependency chains
6. Watch progress in the real-time DAG visualization

Watch It Happen

The dashboard shows everything in real-time:

• Activity feed — live SSE-streamed events as they happen
• Board view — tasks moving across columns as agents work
• Streaming output — live agent output as they think and work
• Thinking blocks — expandable reasoning sections
• Tool calls — every file read, write, bash command, and git operation
• Swarm DAG — parallel execution visualization with dependency edges
• Messaging threads — if configured, watch agents discuss in Slack, Discord, or other connected platforms

Chat With Agents

Talk to any agent directly without going through a project:

1. Go to the Chat page in the dashboard (or use the floating chat widget)
2. Select an agent (e.g., Finn for architecture advice, Grace for meeting transcript processing)
3. Start a conversation — upload files, paste images, or just type

Chat sessions use the same isolated containers and full toolbox.

CLI Access

If you want to interact from the terminal:

pip install djinn-bot-cli

djinn login           # If auth is enabled
djinn chat            # Interactive agent + model selection
djinn status          # Server health
djinn provider list   # See configured providers

Next Steps