A Hermes Swarm comes with 6 preset agent roles out of the box, and picking the right ones is what makes your swarm actually work in production. After running every combination across my own workflows, I'll walk through what each role does, when to use it, and how to combine them for different workflows.
The 6 Preset Roles
When you click "Add swarm" in Hermes Workspace, you can pick from six preset roles — Builder, Reviewer, Triage, Lab, Sage, and Scribeex.
Each comes with preset system prompts and skills, and you can use them out of the box or customise them to fit your workflow.
Role 1 — Builder
Builder is the doer.
What Builder does is draft content, write code, create files, and produce raw output. It's the agent that actually generates the work product.
Use Builder for content creation workflows, code generation, and anything where you need raw output. Best paired with Reviewer to catch the inevitable misses.
Role 2 — Reviewer
Reviewer is the QA agent.
What Reviewer does is read Builder's output, catch errors, suggest improvements, and validate quality. It's the second pair of eyes that turns raw output into shippable work.
Use Reviewer always after a Builder, on any quality-critical workflow, and for production-ready content. The Reviewer pattern is what separates good swarms from broken ones.
I cover the broader critic-step principle (the same pattern as Google Simula's dual critic filter — see Google Simula Mechanism Design) applied here in Hermes.
Role 3 — Triage
Triage is the router.
What Triage does is read incoming requests, decide which agent should handle them, and route work to the right specialist. It's the front door of the swarm.
Use Triage for customer-facing workflows, multi-channel inputs, and when tasks vary widely. Best paired with every other role because Triage acts as the dispatcher.
Role 4 — Lab
Lab is the experimenter.
What Lab does is try new approaches, test hypotheses, and generate options for evaluation. It's the agent that explores rather than executes.
Use Lab for innovation work, A/B testing ideas, and early-stage exploration. Best paired with Sage for evaluation of what Lab generates.
🔥 Want my full Hermes Swarm role playbook? Inside the AI Profit Boardroom, I share my exact Hermes Swarm role configs for content, customer ops, research, and more. Plus 2-hour Hermes course and weekly live coaching. 2,800+ members. → Get the playbook
Role 5 — Sage
Sage is the advisor.
What Sage does is provide strategic input, evaluate options, and offer wisdom on decisions. It's the agent you ask before committing rather than the one that does the committing.
Use Sage for strategic planning, decision support, and high-stakes evaluation. Best paired with Lab for ideating and then evaluating in sequence.
Role 6 — Scribeex
Scribeex is the documentarian.
What Scribeex does is capture what happened, write summaries, and maintain records. It's the historian of the swarm.
Use Scribeex for long workflows where state matters, compliance-sensitive work, and any time you need clear audit trails. Best paired with any combination of others because Scribeex documents what they do.
Common Role Combinations
Here are the four combinations that cover most use cases.
Content creation swarm
Builder drafts the content, Reviewer checks quality, Scribeex keeps records of decisions.
Customer ops swarm
Triage routes incoming messages, Builder drafts responses, Reviewer checks before sending.
Research swarm
Lab explores hypotheses, Sage evaluates findings, Scribeex documents conclusions.
Engineering swarm
Builder writes code, Reviewer reviews PRs, Sage handles architectural decisions.
For each workflow, the right combination matters more than the total number of roles.
Custom Roles
Beyond the 6 presets, you can create custom roles for domain-specific work.
Use custom roles when you're doing legal, medical, or financial work, when you've got unique workflow steps, or when the presets don't quite fit. Click "Add swarm" then "Custom", define the system prompt, set the skills, and save.
Custom roles let you build specialist agents for your specific work without forcing presets to do something they weren't designed for.
Aurora: The Orchestrator
All 6 roles report to Aurora.
Aurora is the master orchestrator who receives your mission, routes work to sub-agents, coordinates handoffs, and reports progress. You don't need to micromanage — you set the goal and Aurora orchestrates the rest.
How To Pick Which Roles For Your First Swarm
If you're new to swarms, start lean and expand as needed.
Start with Builder plus Reviewer for any content or output work. Add Triage if you have varied incoming requests. Add Sage if you want strategic input. Add Lab for experimentation. Add Scribeex for documentation needs.
Don't add all 6 on day one. The complexity ramps up fast and you'll struggle to debug if something breaks.
Time Savings Per Role
Approximate savings from my own use of each combination.
A Builder swarm alone saves about 50% of the time versus solo writing. Builder plus Reviewer saves about 70% and improves quality. Triage plus Builder plus Reviewer saves about 80% on customer ops.
Each role added (when used right) compounds the savings rather than adding them linearly.
Common Role Mistakes
There are four mistakes I see operators make repeatedly with role configurations.
The first is skipping Reviewer. Builder alone produces inconsistent quality. Always add a Reviewer.
The second is adding too many roles. 5+ roles for simple tasks is overkill. Start with 2-3 and expand based on real need.
The third is vague role definitions. If two roles overlap, your master gets confused. Make each role narrow and clear.
The fourth is forgetting handoff rules. Tell each role how to hand off to the next. Without explicit handoffs, agents drift.
Where Roles Limit You
Be honest about the limits.
Roles don't replace specialist domain experts. They don't handle creative judgement well. They don't fully replace senior decision-makers.
For execution work, roles are great. For strategic judgement, humans still matter.
Role Setup Time
Setting up a single role takes 5-10 minutes. Setting up a 3-role swarm takes 20-30 minutes. Setting up a 6-role swarm takes 60-90 minutes.
The time is upfront — then you reuse the configuration repeatedly.
How Roles Pair With Channels
You can connect Hermes Swarm to channels like Telegram and Discord, and there's a common pattern that works.
Triage sits at the channel front door, Builder drafts responses, Reviewer validates before sending. This is the same multi-agent pattern as Telegram AI Agent Architecture but applied in Hermes rather than Telegram-native tooling.
🚀 Want my full role playbook? The AI Profit Boardroom has my Hermes Swarm role templates, OpenClaw 6-hour course, 2-hour Hermes course, daily training, weekly live coaching. 2,800+ members. → Join here
FAQ — Hermes Swarm Roles
Which role should I add first?
Builder — most universal.
Do I need Reviewer?
For quality-critical work, yes. For exploration, optional.
Can custom roles replace presets?
Yes. For specialist domains, custom often fits better.
How does Aurora know which role to call?
Aurora reads role descriptions and your mission, then routes intelligently.
Can two agents have the same role?
Yes. Useful for parallelism — for example, multiple Builders working different parts.
What if a role gets stuck?
Click the agent and check the API and model settings. Usually it's the wrong model assigned.
Can roles share memory?
Limited by Hermes Workspace settings. Configure per agent.
Related Reading
- Hermes Agent Swarm — broader Hermes Swarm overview.
- Hermes Workspace — Workspace overview.
- Hermes Agent Workspace — broader Workspace setup.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
The Hermes Swarm role system is what makes multi-agent workflows actually work — pick the right roles and you've got a real team.