Skip to main content
Menu
Back to projects
Agent Council logo

Agent Council

Formerly Agent Society

AI Mission Control · Hackathon submission

Turn complex briefs into a coordinated council.

Agent Council, formerly Agent Society, is a frontend-first multi-agent AI workspace that turns a mission brief into a focused execution process, readable progress, and a final structured response.

Explore the ProjectWatch walkthrough
Role
Creator and Full-Stack Developer
Status
Hackathon project
Agent Council mission control dashboard with mission brief composer

Brief the council, configure the mission, and launch a visible execution flow.

Meet the council

Nine specialists. One coordinated intelligence.

Each member brings a focused perspective to the mission, joining the council when their expertise can improve the final outcome.

Particle identity · 01

Planner

01 / 09 · orchestrator

Planner

Online
“I turn broad missions into structured workstreams, dependencies and an execution path.”

Role confidence

94%

I focus on

Decomposition, sequencing and coordination.

I contribute: Mission Graphs, task dependencies and execution plans.. I do not: I organise the mission, but I do not replace specialist judgment.. Typical output: Mission Graph.

Typical output

Mission Graph

Project overview

A mission system that shows its reasoning path.

Rather than routing every request through one generic assistant, Agent Council classifies the mission, selects the smallest useful group of specialists, and coordinates their work through a Mission Graph when decomposition is useful.

Mission status

Classifier-led orchestration

The classifier first selects the smallest useful council. Planning, mediation, and synthesis join only when the mission calls for them.

Mission control

Live orchestration dashboard

0 agents active

Demonstration mission

Awaiting classification

Compare the strongest approaches for teaching Python to beginners.

  • Planner

    offline

    Awaiting classification

  • Research Agent

    offline

    Awaiting classification

  • Product Strategist

    offline

    Awaiting classification

  • Technical Architect

    offline

    Awaiting classification

  • Marketing Strategist

    offline

    Awaiting classification

  • Finance Agent

    offline

    Awaiting classification

  • Risk Critic

    offline

    Awaiting classification

  • Mediator

    offline

    Awaiting classification

  • Finalizer

    offline

    Awaiting classification

Classification locked

Research

Direct specialist path

Mission
Classifier
Planner
Research
Product
Architecture
Marketing
Finance
Risk critic
Mediator
Finalizer
Final report

Mission → classifier → selected route → validated report

Execution model

A replayable mission, not a black box.

The mission moves from a brief through planning and three parallel specialist workstreams, then merges into verification and a structured report.

  1. 01 / Intake

    Mission brief

    The operator defines the objective, constraints, and intended outcome.

  2. 02 / Configure

    Mission configuration

    Mission settings shape the council and the operating context.

  3. 03 / Classify

    Mission classification

    The engine selects the smallest useful execution path.

  4. 04 / Plan

    Planning

    The Planner maps dependencies into a visible Mission Graph.

  5. 05 / Execute

    Parallel specialist execution

    Specialists investigate their focused streams at the same time.

  6. 06 / Verify

    Conflict detection

    The council surfaces inconsistencies before they reach the report.

  7. 07 / Resolve

    Mediation

    Conflicting specialist outputs converge into a resolved decision.

  8. 08 / Synthesize

    Final synthesis

    The Finalizer turns validated work into a coherent answer.

  9. 09 / Deliver

    Structured report

    The result remains readable, replayable, and ready to act on.

Three focused specialist paths branch from Execute and converge into Verify before the mission is resolved and delivered.

Mission operating system

More than an answer.
A complete mission workspace.

Agent Council preserves every mission as a configurable, replayable and readable execution record—from the first brief to the final report.

Execution record

Mission replay

Replay mission events from intake through planning, specialist execution, conflict resolution and final synthesis.

Replay timeline

00:00 · Mission started

01 / 08

Mission Control · Brief accepted by Mission Control.

00:00

Local continuity

Mission history

Revisit saved missions, inspect previous outcomes and continue working from earlier execution records.

Stored in this browser

Validated output

Generated reports

Turn validated agent work into a structured report containing decisions, risks, recommendations and next actions.

Validated reportComplete

Before launch

Mission configuration

Shape the council before execution using mission type, depth, agents, timebox and operating preferences.

Mission typeBusiness plan
Execution depthDetailed
Timebox120 sec
Mission modeMission Graph
PlannerResearchFinanceRisk
QualityDetailed

Appearance

Workspace personalization

Adapt the command centre through accent colours, interface preferences, animation controls and accessibility settings.

Mission Control
AccentThemeMotionDensity

Command layer

Keyboard navigation

Move quickly between core workspace views using the application’s direct keyboard shortcuts.

Command barMission Control

Project showcase

Mission control stays legible from planning through final synthesis.

The interface brings mission configuration, workstreams, execution relationships, chapters, and reports into separate focused views rather than exposing raw internal payloads.

The council can suggest a mission setup before the operator launches it.
The workflow view makes task nodes, parallel waves, and synthesis readiness visible.
A network view surfaces the execution relationships between participating specialists.
The timeline turns execution events into a readable story of the mission.
Completed work is brought together into a readable final synthesis and report.

System architecture

One control surface, visible orchestration, local continuity.

The interface gathers a mission brief and configuration, while the Mission Control and Mission Engine classify the work, build or bypass a Mission Graph, coordinate the council, and pass structured results to presentation views. Local browser storage keeps saved settings, mission history, and replay events. Real execution calls a configured Qwen/DashScope-compatible API endpoint.

  • This is a frontend-first hackathon implementation with no bundled backend or database.
  • Mission history and replay data are stored locally in the browser.
  • Qwen calls use the configured Qwen/DashScope-compatible endpoint.
  • The development experience supports mock/demo behavior; real mission execution requires the operator's own API key.
Browser storage preserves settings, history, and replay data; the configured Qwen/DashScope-compatible endpoint enables real mission execution.

Implementation toolkit

  • Next.js
  • React
  • TypeScript
  • Tailwind CSS
  • Qwen/DashScope-compatible API

Project walkthrough

See the council in action.

This walkthrough introduces the mission-control experience and the way the project turns a brief into a visible multi-agent workflow.