Features
Everything in the box. No add‑ons.
One binary. Every feature. No tiers, no paywalls, no plugins to install. Here’s what you get.
01 · WORKFLOWS
Drag nodes. Wire them up. Hit run.
A visual canvas where you build automations by connecting steps. Mix AI agents, API calls, code scripts, and human checkpoints in one flow -- no config files, no code required.
- Build repeatable pipelines by dragging and connecting steps
- Mix agent work, API calls, code scripts, and human approvals in one flow
- Run on a schedule, from a webhook, or with one click
- See every run live -- nodes light up as they execute
- If something crashes, the run picks up where it left off
- Drop a Human Gate node anywhere you want a person to review before continuing
02 · AGENTS
Give each agent a role, a model, and boundaries.
Every agent is a named AI worker with its own personality, tools, and limits. Pick a model, define what files it can touch, and attach skills that teach it your team's conventions. You can spin up a code reviewer that only reads diffs, a designer scoped to /assets, or a researcher with read-only access.
- Choose from Claude, GPT, Codex, Ollama, or any OpenRouter model
- Set allowed and restricted file paths -- enforced automatically
- Three sandbox levels: no access, read-only, or full read/write
- Attach reusable skills to teach conventions and house rules
- Set daily budget caps so spending never surprises you
- Org chart view shows who reports to whom
03 · MISSION CONTROL
Watch every agent. Pause any of them.
A real-time dashboard for your entire swarm. See what each agent is doing, how much it's spending, and shut down anything that goes sideways -- all from one screen.
- See every agent's live output, status, and what they're working on
- Track spending across agents and runs in real time
- Active missions show progress bars -- stuck ones float to the top
- One-click Emergency Pause stops all agents immediately
- Live updates via WebSocket -- no refreshing needed
04 · MISSIONS
Set a goal. Let your agents plan and execute it.
A mission is a bounded goal -- like 'add dark mode' or 'fix the checkout bug.' Your lead agent breaks it into milestones with acceptance criteria, assigns tasks to workers, and a validator agent checks each phase before moving on. If something goes wrong, missions auto-detect when they're stuck and offer one-click recovery.
- Start from a template (Add feature, Fix bug, Write tests, Refactor, Write docs) or describe your own goal
- The lead agent generates a step-by-step plan you can review and edit before starting
- Validator agents check each milestone's acceptance criteria before advancing
- If a mission gets stuck, recover and re-plan with one click
- Mission journal shows a timeline of everything that happened -- every decision, artifact, and handoff
- No mission needed for quick questions -- just chat with an agent or drop a standalone task
05 · TASK BOARD
A kanban board your agents actually use.
Every task flows through a familiar board -- todo, in progress, in review, done. Agents pick up tasks automatically, work them, and move them forward. You can drag cards yourself, set dependencies between tasks, lock files so two agents don't edit the same code, and schedule work for later.
- Drag-and-drop kanban with filtering by agent, priority, project, or status
- Set task dependencies -- agent B won't start until agent A finishes
- File locks prevent two agents from editing the same file at once
- Schedule tasks with start times and soft deadlines
- Comments on tasks auto-inject into the agent's context on next pickup
- Structured handoffs pass summaries and artifacts between agents automatically
06 · APPROVALS
Agents propose. You decide.
Approvals are human checkpoints. Drop a Human Gate into any workflow, or flag a task as needing review. When an agent finishes, the diff lands in your approval queue -- approve to continue, reject to send it back. If you're using worktrees, completed work automatically creates a merge request you can approve with one click.
- Workflow approvals pause a run until you say go
- Task approvals gate work before the next step picks up
- Merge-request approvals show the diff and merge with one click
- Approve or reject from the web UI, Discord, or WhatsApp
- Full history of every approval decision
07 · BRIDGES & CHAT
Talk to your agents like coworkers.
Message any agent from the built-in chat, Discord, or WhatsApp -- the same way you'd Slack a teammate. Agents can also message each other through an internal mailbox to coordinate without you in the loop. Every conversation is saved and searchable.
- In-app chat with streaming responses and full message history
- Discord bridge -- @mention your agent or DM it directly
- WhatsApp bridge -- scan a QR code, start texting your agents
- Agents coordinate through an internal mailbox for handoffs and questions
- Per-channel routing and sender allowlists for security
- Slack and Telegram on the roadmap
08 · PROJECTS & WORKTREES
One workspace per codebase. Git isolation built in.
A project ties your agents to a specific codebase. It sets the working directory, injects context into every agent's prompt automatically, and tracks all activity and spending in one place. Turn on worktrees and each agent gets its own git branch -- no more stomping on each other's edits.
- Browse and edit files, run a terminal, all from the project page
- Every agent automatically gets your project's context -- no re-explaining
- Worktrees give each agent an isolated git branch with auto-sync from main
- Completed work auto-creates merge requests for one-click approval
- Activity feed tracks every task, cost event, and decision in the project
- Set project-level budget caps to control total spending
$ swarmlab skills install code-review
Downloading code-review@1.2.0...
Installed to /skills/code-review
Ready — assign to any agent.09 · SKILLS & CALENDAR
Teach your agents. Schedule their work.
Skills are reusable instructions -- like a style guide, review checklist, or API convention doc -- that you attach to agents so they follow your team's rules every time. The Skill Hub has a pre-built catalog you can enable with one click. And the calendar lets you schedule recurring work with agent reminders, so your standup bot pings the channel every morning without you lifting a finger.
- Create custom skills or browse the built-in Skill Hub catalog
- Attach skills to agents individually or in bulk via skillsets
- Calendar with recurring events (daily, weekly, monthly, custom)
- Calendar reminders auto-spawn tasks and assign them to agents
- Skills auto-inject into CLI agents as files, or into API agents as system messages
- Perfect for standups, weekly reports, quarterly reviews, or any repeating workflow