Get started
FEATURESTREAM E

Agentic Teams

Agentic Teams is QuoxCORE's multi-agent collaboration framework. Define teams of specialist AI agents, submit high-level objectives, and watch them decompose work and execute autonomously - with full budget control and audit trails.

25
Available Agents
4
Topology Modes
314+
Engine Tests
100%
Auditable

Overview

Traditional agent systems run in isolation - one agent, one task, one conversation. Agentic Teams breaks this paradigm by enabling structured collaboration between multiple specialist agents, each with defined roles and boundaries.

A team has an orchestrator that receives high-level objectives and decomposes them into discrete tasks. Each task is routed to the best-suited member agent based on domain expertise. Tasks execute in parallel where possible, with budget guardrails preventing runaway costs.

Single-agent bottleneck
Parallel task execution across specialists
No budget visibility
Token and tool-call budgets with real-time tracking
Manual task decomposition
Orchestrator auto-decomposes objectives into tasks
No scheduled automation
Cron-based triggers with auto-disable on failure

Core Concepts

Six primitives form the foundation of the Agentic Teams system:

Team
A group of AI agents with defined roles, policies, and budgets
Orchestrator
The lead agent that decomposes objectives into tasks
Member
A specialist agent assigned to specific task types
Objective
A high-level goal submitted to the team
Task
An atomic unit of work assigned to a specific agent
Schedule
A cron-based trigger that creates objectives automatically
Topology
How agents coordinate: star (flat), hierarchical (pod leads), pipeline (stages), mesh (collaborative)
Learning Router
Tracks agent performance and routes tasks to the best specialist
Fallback
When an agent fails, a domain-compatible replacement is found automatically

Team Manifest

Every team is defined by a JSON manifest that declares its structure, members, and policies. Manifests can be exported, shared, and installed across QuoxCORE instances.

{
  "id": "secops",
  "type": "team",
  "version": "1.0.0",
  "displayName": "SecOps Team",
  "orchestrator": { "agentId": "sentinel", "domains": ["security"] },
  "members": [
    { "agentId": "nova", "role": "specialist", "domains": ["network"] },
    { "agentId": "cipher", "role": "specialist", "domains": ["cryptography"] }
  ],
  "policies": {
    "maxBudgetTokens": 200000,
    "maxBudgetToolCalls": 100,
    "maxParallelTasks": 3,
    "requireApprovalBeforeStart": true
  }
}
idUnique identifier for the team, used in API calls and routing
typeAlways "team" - distinguishes from agent and pack manifests
versionSemantic version for the team definition
displayNameHuman-readable name shown in the dashboard UI
orchestratorThe lead agent responsible for decomposing objectives into tasks
membersArray of specialist agents with their roles and domain expertise
policiesBudget limits, parallelism caps, and approval requirements

Objectives & Tasks

An objective is a high-level goal submitted to a team. The orchestrator decomposes it into tasks, which are then distributed to member agents. The lifecycle follows a strict state machine:

pending
decomposed
running
completed
1
Objective Created
User or schedule submits a goal
2
Orchestrator Decomposes
Lead agent breaks goal into tasks
3
Tasks Created
Atomic work items assigned to members
4
Tasks Execute in Parallel
Agents work concurrently within budget
5
Objective Completes
All tasks done, results aggregated

Tasks can finish with a status of completed, failed, or skipped. The orchestrator aggregates results and determines the final objective status.

Autonomous Scheduling

Teams can be configured with cron-based schedules that automatically create objectives at defined intervals. This enables unattended operations like nightly security sweeps, hourly health checks, or weekly compliance audits.

Budget Enforcement

Before each scheduled trigger, the system checks remaining daily budget. If the team has exceeded its token or tool-call limits, the trigger is skipped and an alert is raised.

Auto-Disable on Failure

After N consecutive failures (configurable, default 3), the schedule is automatically disabled and the team owner is notified. This prevents runaway error loops from burning budget.

Approval Gates

Schedules respect the team's approval policy. If requireApprovalBeforeStart is true, scheduled objectives enter a pending-approval state rather than executing immediately.

Schedule Presets

Every hour0 * * * *
Daily 9am0 9 * * *
Weekdays 9am0 9 * * 1-5
Weekly Monday 9am0 9 * * 1
Monthly 1st0 9 1 * *

Budget Guardrails

Every team operates within strict budget boundaries. Policies are enforced at the team level and checked before each task execution and scheduled trigger. Exceeding any limit pauses execution and raises an alert.

PolicyDescriptionDefault
maxDailyTokensMax tokens per day for scheduled objectives500,000
maxDailyToolCallsMax tool calls per day500
maxUnattendedObjectivesMax auto-triggered objectives per day10
alertAtBudgetPercentAlert when this % of budget is used80%
canStartWithoutApprovalAuto-approve scheduled objectivesfalse

Budget tracking is per-calendar-day (UTC). Counters reset at midnight. Real-time usage is visible in the team dashboard under the Budget tab, with a progress bar and projected daily burn rate.

Team Packs

A Team Pack is a bundle of related teams that can be installed together in a single operation. Packs define which teams are included, their default configurations, and any shared policies. This makes it easy to deploy pre-configured multi-team setups.

{
  "id": "quoxcore-demo-pack",
  "type": "pack",
  "version": "1.0.0",
  "displayName": "QuoxCORE Demo Pack",
  "description": "Starter pack with QuoxCORE and SecOps teams",
  "teams": [
    { "manifestUrl": "./teams/quoxcore.json" },
    { "manifestUrl": "./teams/secops.json" }
  ],
  "sharedPolicies": {
    "maxDailyTokens": 1000000,
    "alertAtBudgetPercent": 75
  }
}

Built-in: QuoxCORE Demo Pack

Ships with every QuoxCORE instance. Includes the quoxcore general-purpose team and the secops security operations team. A great starting point for learning the Agentic Teams system.

Custom Packs

Create your own packs by bundling team manifests. Packs can be exported as JSON and shared across QuoxCORE instances via the /api/teams/install-pack endpoint or the dashboard UI.

Team rooms (shipped 2026-07)

Every agentic team gets its own governed Matrix room, created when the team is installed or created and removed when the team is deleted. The room holds the team's orchestrator, its member agents, and your org's people, so a human can watch a team work or step in without leaving Rooms for a separate chat tool.

Replies in a team room are governed and witnessed the same way as any other Quox comms surface: every turn goes through the collector and is recorded in the WARD witness chain, not a side channel the team can bypass.

To provision a room for a team that doesn't have one yet, run quox team provision-room <teamId> from the CLI.

API Reference

All team operations are available through the QuoxCORE REST API:

MethodPathDescription
GET/api/teamsList installed teams
POST/api/teamsCreate custom team
GET/api/teams/:idGet team details
PUT/api/teams/:idUpdate team
DELETE/api/teams/:idUninstall team
POST/api/teams/install-packInstall a team pack
GET/api/teams/:id/exportExport team manifest
GET/api/objectivesList objectives
POST/api/objectivesCreate objective
GET/api/schedules/:teamIdList schedules
POST/api/schedules/:teamIdCreate schedule

Example: Create an Objective

POST /api/objectives
Content-Type: application/json

{
  "teamId": "secops",
  "title": "Nightly security sweep",
  "description": "Scan all fleet hosts for open ports, expired certs, and CVEs",
  "priority": "normal",
  "tags": ["security", "compliance"]
}