Functions

A guide to using SwarmOS: key features, setup, and how to get the most out of your AI team.

Getting Started

Sign Up & First Task

Sign up at app.swarmos.tech
Create your first task from the web UI, or connect a channel (Telegram, Slack) and create tasks from there
Tasks move through: Todo → In Progress → In Review → Done

Connecting Channels

SwarmOS delivers notifications and approvals to the channels you connect: Telegram, Slack, Discord, WhatsApp, or Microsoft Teams. See Integrations for setup.

Key Functions

Create Tasks

Create tasks from:

Web UI — Task board, "+ New Task" button
Telegram — /run <agent> <task>
Slack / Discord — Chat command or web UI
Chat — Natural language: "Create a task to fix the login bug"

Assign Agents

Tasks are assigned to agents based on skills and availability. SwarmOS includes coordinator, development, content, and support agents. Assign manually or let the coordinator auto-route.

Review & Approve

When a task reaches In Review:

Web UI — Approve, request changes, or implement fixes
Telegram — Inline buttons
Slack — Interactive buttons; modal for custom instructions
Discord — Approve, Changes, Reject buttons

Implement Fixes

Re-run the agent with review findings and optional custom instructions. Works from UI, Telegram (/fix <task> -- <instructions>), Slack (modal), or chat.

Setup Steps

Integrations — Connect Telegram, Slack, Discord, WhatsApp, or Teams in Settings
LLM Provider — Choose a provider and add your API key in Settings (see providers)
Skills — Agents use skills from the catalog (see skills)
Linear / HubSpot — Optional: sync tasks and contacts

Task Workflow

Todo → In Progress → Testing → In Review → Done
                ↑                    │
                └────────────────────┘
           (Implement Fixes / Request Changes)

Quality checks (type check, tests, lint, build) run during the testing phase.

Auto-Iterate

Agents can auto-retry tasks when quality gates fail—up to a configurable max (e.g. 3). Flow: execute → run quality gates → if gates fail: retry with gate output as feedback → repeat until max or pass → then human review.

Configure per-task (Create modal: Auto-Iterate toggle + max iterations) or per-agent (Agent Manager > Config: Auto-Iterate section). Task-level overrides agent-level. In the task detail modal you see the iteration counter (e.g. "2/3"), a Stop button, and gate results.

System	Triggers On	Use Case
Auto-Iterate	Success + gate failure	"Works but not good enough"
Escalation	Errors/crashes	Task keeps failing
Persistent	Always	Long-running loops
Implement Fixes	Manual	Human-directed retry

Chat Commands

Use natural language in the web chat, Telegram, or Slack. The AI interprets commands such as: create task, search tasks, update status, implement findings (with optional custom instructions), start crawl.