15 free templates

Agent Teams Workflow Templates

These free multi-agent workflow templates put a team of AI agents to work. One agent leads. The rest research, write, and check the results. They run on Heym, a free platform you host yourself. You build each team on a visual canvas, with no code to write.

researcher-writer teams
validator agents
nested agents
specialist routing

Browse templates

Use cases

What you can automate with Agent Teams templates

Orchestrator and sub-agents

One agent hands work to a research agent and a writer agent. It then returns a clean draft.

Researcher and writer teams

Split research and writing across agents so each one has a clear job.

Validator loops

Add an agent that checks the output before it is final.

Specialist routing

Send each task to the agent that fits it best.

Popular Agent Teams templates

Orchestrator → Research & Writer Agents

An Orchestrator agent hands the topic to a Research agent and a Writer agent. The result is a finished draft for newsletters, blog posts, or digests.

Browse all Agent Teams templates

Pick a workflow, copy the JSON, and adapt the nodes for your team.

15 templates·Agent Teams

Sticky Note

Orchestrator → Research & Writer Agents

Orchestrator agent receives the topic, delegates to ResearchAgent (facts) and WriterAgent (prose).

Main path

Topic → Orchestrator → ResearchAgent → WriterAgent → Draft

Topic or brief

Orchestrator

GLM-4.7

ResearchAgent

sub-agent

GLM-4.7

WriterAgent

sub-agent

GLM-4.7

Newsletter draft

Orchestrator → Research & Writer Agents

Agent Teams

An Orchestrator agent hands the topic to a Research agent and a Writer agent. The result is a finished draft for newsletters, blog posts, or digests.

#Agent#Orchestrator#Multi-Agent

Sticky Note

Before you run

Add Slack credentials to the Trigger node.
Add OpenAI (or compatible) credentials to all three Agent nodes.
Verify channel names before sending live traffic.

Sticky Note

Slack Orchestrator + Triage Sub-Agents

Orchestrator decides severity; UrgentAgent or NormalAgent handles posting.

Main path

InboxMessages → Orchestrator → IsUrgent → UrgentAgent → AlertUrgent

InboxMessages

Orchestrator

GLM-4.7

UrgentAgent

sub-agent

GLM-4.7

NormalAgent

sub-agent

GLM-4.7

AlertUrgent

QueueNormal

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Double-check AlertUrgent before sending alerts or messages to a live destination.

switch

urgent

normal

default

Slack Orchestrator + Triage Sub-Agents

Customer Support

Orchestrator agent classifies incoming Slack messages and delegates to an UrgentAgent or NormalAgent sub-agent, each with its own posting logic.

#Slack#Orchestrator#Multi-Agent

Sticky Note

System Overview

Adversarial code review system. User provides code change (diff/PR), system returns structured review.

How It Works

1. architectAgent: Orchestrator. Classifies change, calls 3 specialists in parallel.
2. reviewerAgent: Detects framing tensions between PR description and diff.
3. implementerAgent: Generates test code to verify claims.
4. reasonerAgent: Identifies high-leverage review angles for the change.

Output Format

VERDICT: approve | request_changes | discuss
CHANGE_CLASSIFICATION: Change type
FRAMING_NOTES: reviewerAgent findings
CONCERNS: Detected issues list
REVIEW_FOCUS: Review focus areas

Sticky Note

Agent Roles

architectAgent (Orchestrator)

Role: Orchestrator and decision maker
Task: Analyzes input, classifies change, calls 3 specialists, integrates results.
Note: Does not review itself, only coordinates.

reviewerAgent (Adversarial Reviewer)

Role: Adversarial critic
Task: Detects framing tensions between PR description and diff.
Harness: anti-deception - Maintains honesty under pressure
Output: CONCERN or NO_CONCERN + severity

implementerAgent (Verification Engineer)

Role: Code verifier
Task: Generates minimal test code to verify change claims.
Harness: code - Engineering constraints and failure modes
Output: VERIFICATION code + CLAIM + SEVERITY_HINT

reasonerAgent (Review Angles)

Role: Analytic decomposer
Task: Identifies which angles to scrutinize in this change.
Harness: reasoning - Causal analysis and risk assessment
Output: REVIEW_ANGLES list (ranked by leverage)

Sticky Note

Harness Integration

What is Harness?

Each agent receives a cognitive injection — constraints, procedures, and failure-mode blockers that shape their thinking.

Three Harness Types

1. ejentumAntideception (HTTP Node)

Mode: anti-deception
User: reviewerAgent
Purpose: Blocks sycophancy, hallucination, manipulation risks
139 abilities across 6 domains

2. ejentumCode (HTTP Node)

Mode: code
User: implementerAgent
Purpose: Prevents bugs, security issues, architectural problems
128 abilities across 13 disciplines

3. ejentumReason (HTTP Node)

Mode: reasoning
User: reasonerAgent
Purpose: Causal analysis, risk assessment, tradeoff evaluation
311 abilities across 6 dimensions

Flow

1. Agent delegated → First action: Harness call
2. Harness API returns injection → Agent absorbs it
3. Agent executes task with injection active
4. Agent verifies output against injection constraints

Rule: Skipping harness call = Acting like base model = Defeating delegation purpose

chatInput

architectAgent

anthropic/claude-3.7-sonnet:thinking

ejentumAntideception

ejentumCode

ejentumReason

reviewerAgent

sub-agent

google/gemini-2.5-pro

implementerAgent

sub-agent

z-ai/glm-5.1

reasonerAgent

sub-agent

qwen/qwen3-max-thinking

Ejentum Code Reviewer

Dev & IT Ops

A 4-agent MAS that refuses to rubber-stamp PRs: architect orchestrates three cross-lab specialists (reasoning, code, anti-deception), each backed by the Ejentum Logic API.

#Multi-Agent#Code Review#AI

start

steelmanAgent

openai/gpt-5-nano

stresstestAgent

anthropic/claude-opus-4

gapfinderAgent

z-ai/glm-4.7

setFields

Sticky Note

Blind Eval Trio
Pre-commitment self-evaluation for agent runtimes.

3 cross-lab agents → 3 independent evaluations:
steelman (defends), stress_test (attacks), gap_finder (what's missing).
No synthesizer — calling agent integrates.

Structural integrity
• 3 model labs → cross-lab decorrelation
• Tool lockout per role (no role-switching)
• No synthesizer agent (3 raw outputs)
• Forced output structure + articulation gate

steelman
gpt-5-nano + harness_reasoning
Role: strongest case FOR the method.
Hard rule: pure advocacy. Zero smuggled critique.

stress_test
claude-opus-4 + harness_anti_deception

Role: where the method BREAKS.
Hard rule: severity-tagged failures + concrete scenarios.

gap_finder
glm-4.7 + harness_memory

Role: what's MISSING (steps + articulation depth).
Hard rule: name 3 deeper implicit assumptions.

Output: structured JSON
{ steelman, stress_test, gap_finder, usage_note }

Three voices, no synthesis. Integrate; do not aggregate.

Sticky Note

Verified across domains

Caught hidden behavior changes (refactor planning), compliance gaps (payments migration), premature containment (security IR), framing tensions (investigative cases), unvalidated assumptions (strategic decisions including meta-eval of itself).

Role discipline + cross-lab decorrelation generalize. No domain-specific tuning required.

Sticky Note

Example output

Task: Auto-merge Dependabot PRs after CI passes.

steelman defends:
"Leverages standard platform capabilities; CI gating preserves code quality; uniform across all updates."

stress_test attacks (HIGH):
"Dependency hijacking — compromised maintainer publishes backdoored version, CI passes (tests don't catch malware payload), auto-merge deploys to production."

gap_finder finds:
"Missing: scope limiting, version constraints, rollback procedure. 'CI catches breaking changes' silently assumes integration coverage AND env parity AND end-to-end scenarios."

→ One agent advocates, one attacks, one finds gaps. Calling agent integrates.

Sticky Note

Callable from any agent

heym exposes this workflow as an HTTP endpoint:
POST /api/workflows/{id}/execute/stream

SSE events stream as each node completes. Any agent runtime ;
Claude Code, Cursor, Codex, custom Python — can curl it directly. One workflow, infinite integrations.

Sticky Note

Built with heym + ejentum-mcp cognitive harness.
Repo:
github.com/ejentum/agent-teams
ejentum.com

Blind Eval Trio

AI Agents

Three cross-lab agents evaluate any decision blind: steelman defends, stress_test attacks, gap_finder finds what's missing. No synthesizer — you integrate.

#Multi-Agent#Evaluation#Ejentum

start

triageOrchAgent

z-ai/glm-5.1

interpreterAgent

sub-agent

qwen/qwen3-max-thinking

doctorpushAgent

sub-agent

anthropic/claude-opus-4

differentialAgent

sub-agent

deepseek/deepseek-r1

fetchPubmedSearch

fetchNihConditions

fetchNihLabTests

Sticky Note

Blood Panel Triage Team
Patient-education triage for blood lab results.
NOT a diagnostic tool. NOT a replacement for a clinician.
triageOrchAgent (glm-5.1)
Deterministic safety gate (12-marker panic table).
Critical → EMERGENCY OUTPUT, stops.
Abnormal/normal → fans out 3 sub-agents in parallel, integrates.
interpreterAgent (qwen3-max-thinking)
harness_reasoning + PubMed. Plain-language explainer per marker.
doctorpushAgent (claude-opus-4)
harness_anti_deception + NIH lookups. Blunt questions to push the doctor on. No false reassurance.
differentialAgent (deepseek-r1)
3-5 conditions consistent with the pattern, each with confirm/rule-out trio.
Cross-lab diversity: Anthropic, Alibaba, Zhipu, DeepSeek.
Built with heym + ejentum-mcp.
github.com/ejentum/agent-teams · ejentum.com

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Confirm the demo URLs and selectors still match the endpoint you want to monitor.

Blood Panel Triage Team

Healthcare

Orchestrator runs a deterministic 12-marker panic-value gate, then fans out three cross-lab sub-agents to interpret labs, push doctor questions, and enumerate differentials.

#Multi-Agent#Medical#Triage

Sticky Note

Setup

1. Add an LLM credential to orchestrator, primaryReviewer, and challenger — mix vendors so the challenger differs from the reviewer.
2. The GitHub MCP reads your token from the global variable githubPat (Settings → Global Variables).
3. Replace your-github-username in the orchestrator prompt with your handle.

prInput

orchestrator

zai-glm-4.7

primaryReviewer

sub-agent

anthropic/claude-opus-4.8

challenger

sub-agent

openai/gpt-5.5

finalReview

Adversarial PR Review

Dev & IT Ops

An orchestrator pulls a pull request from GitHub, a primary reviewer proposes findings, and a challenger on a different model disputes them — arbitrated into one calibrated approve, comment, or request-changes decision.

#Multi-Agent#Adversarial#Code Review

Sticky Note

Market research team

Orchestrator → MarketAnalyst + CompetitorAnalyst → ResearchSynthesizer.

Add an LLM credential to all four Agent nodes.
Each agent sees only what edges and expressions pass.
Validate the brief; do not treat it as final truth.

marketBrief

Orchestrator

GLM-4.7

MarketAnalyst

sub-agent

GLM-4.7

CompetitorAnalyst

sub-agent

GLM-4.7

ResearchSynthesizer

sub-agent

GLM-4.7

researchBrief

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

Market Research Agent Team

Agent Teams

An orchestrator coordinates a market analyst, a competitor analyst, and a synthesizer to produce a research brief with opportunities and risks.

#Multi-Agent#Orchestrator#Market Research

Sticky Note

Feedback triage team

Orchestrator → SentimentAnalyst + IssueCategorizer → ResponsePlanner.

Add an LLM credential to all four Agent nodes.
Paste feedback, one item per line.
The report guides the team; it is not an auto-decision.

feedbackBatch

Orchestrator

GLM-4.7

SentimentAnalyst

sub-agent

GLM-4.7

IssueCategorizer

sub-agent

GLM-4.7

ResponsePlanner

sub-agent

GLM-4.7

triageReport

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

Customer Feedback Triage Team

Agent Teams

An orchestrator runs a sentiment analyst, an issue categorizer, and a response planner to triage a batch of customer feedback.

#Multi-Agent#Orchestrator#Customer Feedback

Sticky Note

Trip planner team

Orchestrator → DestinationResearcher + BudgetPlanner → ItineraryWriter.

Add an LLM credential to all four Agent nodes.
Costs are estimates, not live prices.

tripBrief

Orchestrator

GLM-4.7

DestinationResearcher

sub-agent

GLM-4.7

BudgetPlanner

sub-agent

GLM-4.7

ItineraryWriter

sub-agent

GLM-4.7

tripPlan

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

Trip Planner Agent Team

Agent Teams

An orchestrator runs a destination researcher, a budget planner, and an itinerary writer to produce a day-by-day trip plan within budget.

#Multi-Agent#Orchestrator#Travel

Sticky Note

Content production team

Orchestrator -> ContentResearcher + DraftWriter -> EditorPublisher.

Add an LLM credential to all four Agent nodes.
Each agent sees only what edges and expressions pass.
Fact-check the article; it is a draft, not final truth.

contentBrief

Orchestrator

GLM-4.7

ContentResearcher

sub-agent

GLM-4.7

DraftWriter

sub-agent

GLM-4.7

EditorPublisher

sub-agent

GLM-4.7

finalArticle

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

Content Production Agent Team

Agent Teams

An orchestrator runs a researcher, a draft writer, and an editor to turn a topic brief into a polished, publish-ready article.

#Multi-Agent#Orchestrator#Content

Sticky Note

Incident response team

Orchestrator -> IncidentTriager + RootCauseAnalyst -> CommsRemediationLead.

Add an LLM credential to all four Agent nodes.
Paste the alert, symptoms, and recent changes.
Output is decision support, not auto-remediation.

incidentSignal

Orchestrator

GLM-4.7

IncidentTriager

sub-agent

GLM-4.7

RootCauseAnalyst

sub-agent

GLM-4.7

CommsRemediationLead

sub-agent

GLM-4.7

responsePack

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

Incident Response Agent Team

Agent Teams

An orchestrator runs a triage analyst, a root cause analyst, and a comms lead to turn an alert into a triage report, likely cause, and status update.

#Multi-Agent#Orchestrator#Incident Response

Sticky Note

RFP response team

Orchestrator -> RequirementsParser + SolutionWriter -> ComplianceReviewer.

Add an LLM credential to all four Agent nodes.
Paste the RFP text or key requirements.
Verify answers against real product capabilities.

rfpInput

Orchestrator

GLM-4.7

RequirementsParser

sub-agent

GLM-4.7

SolutionWriter

sub-agent

GLM-4.7

ComplianceReviewer

sub-agent

GLM-4.7

rfpDraft

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

RFP Response Agent Team

Agent Teams

An orchestrator runs a requirements parser, a solution writer, and a compliance reviewer to draft a structured first-pass RFP response.

#Multi-Agent#Orchestrator#RFP

Sticky Note

Due diligence team

Orchestrator -> FinancialAnalyst + LegalRiskAnalyst -> DiligenceSynthesizer.

Add an LLM credential to all four Agent nodes.
Enter the deal context and known facts.
Verify every finding against primary documents.

dealInput

Orchestrator

GLM-4.7

FinancialAnalyst

sub-agent

GLM-4.7

LegalRiskAnalyst

sub-agent

GLM-4.7

DiligenceSynthesizer

sub-agent

GLM-4.7

diligenceBrief

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

Due Diligence Agent Team

Agent Teams

An orchestrator runs a financial analyst, a legal and risk analyst, and a synthesizer to produce a structured due diligence brief.

#Multi-Agent#Orchestrator#Due Diligence

Sticky Note

Hiring committee team

Orchestrator -> ResumeScreener + SkillsAssessor -> DecisionSynthesizer.

Add an LLM credential to all four Agent nodes.
Paste role requirements and a candidate summary.
Output is decision support, not an auto-decision.

candidateInput

Orchestrator

GLM-4.7

ResumeScreener

sub-agent

GLM-4.7

SkillsAssessor

sub-agent

GLM-4.7

DecisionSynthesizer

sub-agent

GLM-4.7

hiringRecommendation

Sticky Note

Live run checklist

Run once with sample data and inspect each node output in sequence.
Replace sample values early so downstream expressions are validated with real data.
Keep node labels stable if you update expressions that reference earlier outputs.

Hiring Committee Agent Team

Agent Teams

An orchestrator runs a resume screener, a skills assessor, and a decision synthesizer to produce a structured, criteria-based hiring recommendation.

#Multi-Agent#Orchestrator#Hiring

Sticky Note

Before you run

Replace the sample Agent Card discovery URL.
Connect Redis for the previous fingerprint.
Connect an LLM credential and Discord.
Run once to seed the initial fingerprint.
Add separate end-to-end task tests for critical agents.

AgentCardSchedule

FetchRemoteAgentCard

BuildAgentCardFingerprint

LoadPriorAgentCard

AgentCardChangedGate

true

false

ExplainAgentCardChange

z-ai/glm-4.7

SaveAgentCardFingerprint

PostA2aMigrationBrief

AgentCardChangeReceipt

AgentCardSteadyReceipt

Sticky Note

Live run checklist

Test both exits around AgentCardChangedGate before going live.
Tune the schedule, wait, or batch size first so repeated runs stay predictable.
Double-check PostA2aMigrationBrief before sending alerts or messages to a live destination.

A2A Agent Card Compatibility Monitor

Agent Teams

Watch a remote A2A Agent Card for capability or authentication changes and publish a migration brief when the contract drifts.

#A2A Protocol#Agent2Agent#Agent Card

All 15 Agent Teams templates, A to Z

Why teams run Agent Teams workflows on Heym

Heym is a source-available, self-hosted AI workflow platform for agentic automation: agents that call tools, retrieve from your own documents, and pause for human review when a step needs it.

Free and source-available

Released under MIT with a Commons Clause, with no per-task or per-execution pricing.

Self-hosted

Runs on your own infrastructure with Docker, so your data and credentials stay yours.

One-paste import

Copy a template JSON, paste it onto the canvas, and run it in minutes.

Agentic by default

Agent nodes, MCP tools, and built-in RAG on a visual workflow canvas.

Agent Teams templates FAQ

What is a multi-agent system?

A multi-agent system splits one job across a few focused agents. A lead agent guides them. The Orchestrator with Research and Writer Agents template shows this. It turns one request into a finished draft.

Can I customize a template after importing it?

Every node, prompt, credential, and connection is editable after import. Swap the model, change the trigger, add or remove tools, and rewire branches directly on the canvas without writing boilerplate.

Is Heym free and self-hosted?

Yes. Heym is source-available under an MIT license with a Commons Clause, so you can run every template on your own infrastructure with Docker. There is no per-task or per-execution pricing, and your data and credentials stay inside your environment.

What makes these different from ordinary automation templates?

Each one runs on an AI-native canvas, so agent nodes, MCP tools, and built-in RAG sit in the same workflow as the triggers, branching, and integrations. That is the difference between a fixed sequence of steps and an agentic workflow that can retrieve context, call tools, and pause for human review while it runs.

Multi-agent AI systems LLM orchestration Prompt chaining

Template categories

Explore other use cases

View all templates

Marketing & SEO

Content repurposing, social publishing, SEO research, analytics reporting, and audience monitoring for growth teams.

Sales & CRM

Lead qualification, CRM enrichment, personalized outreach, follow-up workflows, and revenue handoffs.

Customer Support

Email triage, FAQ replies, human approval checkpoints, inbox summaries, and customer escalation routing.

Legal & Compliance

Governed research, review checkpoints, policy-sensitive document handling, and audit-friendly agent workflows.

HR & Recruiting

Candidate search, resume-style ranking, recruiting summaries, interview preparation, and hiring pipeline automation.

Finance Ops

Invoice extraction, payment webhooks, receipt creation, pricing monitors, and finance data handoffs.

Document Ops

RAG ingestion, document Q&A, PDF and DOCX workflows, translation, file sharing, and structured extraction.

Data & Analytics

Spreadsheet enrichment, ETL-style syncs, schema remapping, BigQuery handoffs, and reporting pipelines.

Dev & IT Ops

Release monitoring, Sentry issue triage, API reliability, code review, WebSocket alerts, RabbitMQ jobs, and internal tool workflows.

App Integrations

Connect Slack, Discord, Telegram, HTTP APIs, MCP tools, and external systems with reusable workflow patterns.

Personal Productivity

Personal assistants, briefing emails, meeting summaries, lightweight research, reminders, and daily automation.

Education & Learning

Language learning, study guides, lesson planning, assessment feedback, research workflows, and LMS operations.

Healthcare

Clinical-style triage examples, structured health summaries, specialist review patterns, and careful human oversight.

Logistics & Supply Chain

Shipment tracking, delivery exceptions, carrier rates, freight audit, dock scheduling, and supply chain data checks.

Energy & Utilities

Smart meter anomalies, field work orders, outage updates, solar performance, utility bill checks, and interconnection reviews.

E-commerce

Return triage, review insights, order fraud screening, cart recovery, product catalog quality, and inventory checks.

AI Agents

General LLM and agent workflows for classification, generation, evaluation, image creation, and text-to-workflow examples.

Operations Automation

Scheduled checks, debounce steps, generic routers, async dispatchers, and repeatable business operations.

Agent Skills

Agent templates with bundled skills, custom tools, Python actions, and repeatable expert workflows.

What you can automate with Agent Teams templates

Orchestrator and sub-agents

Researcher and writer teams

Validator loops

Specialist routing

Popular Agent Teams templates

Orchestrator → Research & Writer Agents

Browse all Agent Teams templates

Orchestrator → Research & Writer Agents

Slack Orchestrator + Triage Sub-Agents

Ejentum Code Reviewer

Blind Eval Trio

Blood Panel Triage Team

Adversarial PR Review

Market Research Agent Team

Customer Feedback Triage Team

Trip Planner Agent Team

Content Production Agent Team

Incident Response Agent Team

RFP Response Agent Team

Due Diligence Agent Team

Hiring Committee Agent Team

A2A Agent Card Compatibility Monitor

Why teams run Agent Teams workflows on Heym

Free and source-available

Self-hosted

One-paste import

Agentic by default

Agent Teams templates FAQ

Related guides

Explore other use cases

Marketing & SEO

Sales & CRM

Customer Support

Legal & Compliance

HR & Recruiting

Finance Ops

Document Ops

Data & Analytics

Dev & IT Ops

App Integrations

Personal Productivity

Education & Learning

Healthcare

Logistics & Supply Chain

Energy & Utilities

E-commerce

AI Agents

Operations Automation

Agent Skills

Orchestrator → Research & Writer Agents

Slack Orchestrator + Triage Sub-Agents

Ejentum Code Reviewer

Blind Eval Trio

Blood Panel Triage Team

Adversarial PR Review

Market Research Agent Team

Customer Feedback Triage Team

Trip Planner Agent Team

Content Production Agent Team

Incident Response Agent Team

RFP Response Agent Team

Due Diligence Agent Team

Hiring Committee Agent Team