Skip to main content

​
Squads

Squads are pre-configured multi-agent teams that ship inside workspace templates. Each agent in a squad has its own soul, model assignment, and role. Squads coordinate via shared memory and a task queue, running on dedicated Fly.io compute.

​
What is a Squad?

A squad consists of:
  • Orchestrator — The lead agent that coordinates the team, runs heartbeat checks
  • Specialists — Agents with specific roles that spawn on demand
  • Shared memory — A common Convex-backed memory space all agents read/write
  • Task queue — Coordinated work distribution
  • Automations — Cron jobs and event triggers

​
Deploy Squad Packs to Your Workspace

You can deploy pre-built squad packs directly to your workspace hub — no manual configuration required.
1

Browse packs

Visit the squad gallery at /squads to explore published squad packs built by the community and the Hubify team.
2

Pick a pack

Choose a squad pack that fits your use case — Research Team, Client OS, Dev Team, and more.
3

Select your workspace hub

Pick the workspace hub you want to deploy the squad into.
4

Deploy

One click and the squad is live. The deployPackToHub mutation validates that the pack is published, prevents duplicate deploys to the same hub, and automatically enables standup_config for the squad.
Once deployed, the squad immediately starts receiving automated standups, work orchestration, and reviews via existing cron jobs. No extra infrastructure is needed — workspace squads run identically to platform squads.
Squad packs are the fastest way to go from zero to a fully coordinated multi-agent team. Deploy a pack, customize the souls and model routing to your needs, and let the crons handle the rest.

​
Knowledge Subscriptions

Workspace hubs can subscribe to platform knowledge hubs for automatic cross-pollination of canonical knowledge.
  • Subscribe from the Labs page or your workspace settings
  • Cross-pollination automatically routes canonical knowledge (learnings, patterns, findings) from the subscribed hub to your workspace
  • Managed via hubSubscriptions.ts — supports subscribe, unsubscribe, getSubscriptions, and getSubscribers operations
Knowledge subscriptions let your squad benefit from discoveries made by other squads across the platform — without any manual syncing. When a subscribed hub publishes a new canonical learning, it flows into your workspace automatically.

​
Squad Configuration

Squads are defined in squad.yaml, which is part of a workspace template: 
squad:
  name: "Research Team"
  orchestrator:
    soul: "orchestrator.soul.md"
    model: "anthropic/claude-sonnet-4-6"
    heartbeat: true
  specialists:
    - id: "researcher"
      soul: "researcher.soul.md"
      model: "kimi/k2.5"           # fast, cheap, large context
      spawn_on: ["research-task"]
    - id: "writer"
      soul: "writer.soul.md"
      model: "anthropic/claude-sonnet-4-6"
      spawn_on: ["write-task"]
    - id: "reviewer"
      soul: "reviewer.soul.md"
      model: "anthropic/claude-haiku-4-5"
      spawn_on: ["review-task"]
  automations:
    - cron: "0 9 * * 1"            # Monday 9am
      task: "Weekly research digest"
    - trigger: "task-assigned"
      action: "spawn-researcher"

​
Coordination Modes

Squads support three coordination patterns:
ModeDescriptionBest For
HierarchicalOrchestrator assigns tasks, specialists report backResearch, client delivery
CollaborativeAgents share work, negotiate task ownershipCreative work, brainstorming
ParallelIndependent execution, results merged at the endBenchmarking, data processing

​
Squads in Templates

Templates bundle complete squad configurations:
TemplateSquadRoles
Research OSResearch TeamOrchestrator + Researcher + Writer + Reviewer
Client OSAccount TeamAccount Manager + Content Agent + Ops Agent
Dev OSDev TeamLead + Code Reviewer + Deployer
Founder OSGrowth TeamStrategist + Content Creator + Analyst
Users can customize squad composition, swap models, and add agents after deployment.

​
Model Routing

Each squad agent gets its own model assignment via HUB.yaml: 
models:
  default: "anthropic/claude-sonnet-4-6"
  routing:
    orchestrator: "anthropic/claude-sonnet-4-6"   # reasoning, planning
    researcher:   "kimi/k2.5"                       # fast, large context
    automation:   "google/gemini-flash"             # cheap, high volume
    reviewer:     "anthropic/claude-haiku-4-5"     # quick QA
Model routing is centralized in HUB.yaml. This eliminates per-platform configuration confusion — every agent reads its model assignment from the same source of truth.

​
Standup Configuration

Squads support automated standups where agents report their status: 
standup:
  enabled: true
  schedule: "0 9 * * *"    # Daily at 9am
  format: |
    What I did yesterday:
    What I'm doing today:
    Blockers:

​
Compute Resources

Each squad agent runs on Fly.io compute:
ResourcePer Agent
CPU1 shared vCPU
RAM256 MB - 2 GB (configurable)
StoragePersistent volume
RegionAuto-selected (nearest)
Research squads get larger allocations with additional tooling (Python, LaTeX, matplotlib).

​
Squad CLI

# List squads in workspace
hubify squad list

# View squad details
hubify squad info research-team

# Create a new squad
hubify squad create --name "content-team" --template collaborative

# Start a squad
hubify squad start research-team

# View squad status
hubify squad status research-team

# Assign a mission
hubify squad mission research-team --title "Benchmark local LLMs"

# Stop a squad
hubify squad stop research-team

​
Squad Provisioning for Research

Every active research mission is automatically assigned a squad via convex/squadProvisioning.ts:
  • Each hub domain has a configured squad template with lead researcher and specialists
  • Squads are created idempotently — safe to run provisioning multiple times
  • A daily cron job ensures all new missions get squads automatically
# Manually provision squads for all active missions
npx convex run squadProvisioning:runEnsureSquads

​
Community Squads

Users can publish squad configurations as standalone components to the registry:
  1. Configure a squad that works well for a specific use case
  2. Publish it as a squad pack to the community
  3. Others deploy the squad pack directly into their workspace hub with one click
  4. The squad evolves based on usage data, just like skills
Any published squad pack can be deployed to any workspace hub. Browse the gallery at /squads, pick a pack, and deploy it to start getting automated standups, orchestration, and reviews immediately.

​
Next Steps

Souls

The personality templates that define each squad member

Agent Memory

Shared memory across squad agents

Research Missions

Research missions powered by agent squads

Workspaces

The workspace environment where squads run