multi-agent-ff15

OpenCode Multi-Agent Web Dashboard

Manage and monitor your AI agent team from a browser

Noctis Team screen — dispatch missions and monitor agent activity in real time

---

multi-agent-ff15 is the current browser-first application for FF15-inspired OpenCode workflows. It replaces the older desktop/Tauri app in this repository and turns a tmux-heavy multi-agent setup into a single browser surface for dispatching work, monitoring sessions, reviewing reports, and inspecting runtime state.

Built on: OpenCode · React Router 7 · TypeScript

Why Use It

• Run Noctis-led multi-agent workflows from a browser instead of living in tmux.
• See what agents actually did through live sessions, tool traces, and reports.
• Keep missions, projects, configuration, MCP state, and runtime health in one place.

Overview

The dashboard is organized around a human workflow rather than raw runtime internals:

• Noctis Team: start work and watch party activity
• OpenCode Sessions: inspect session history, tool calls, models, and context/execution selection
• Reports: review long-form output after a mission finishes
• Projects: choose where work runs and launch the right workspace
• Config: edit app settings and bootstrap required local config
• MCP and Server: inspect MCP configuration and OpenCode runtime state

Typical Flow

1. Send a mission from Noctis Team.
2. Watch the party respond and inspect detailed session activity.
3. Open Reports when you need a cleaner long-form result.
4. Use Projects, Config, MCP, or Server when you need execution context or runtime debugging.

Agent Relationships

flowchart TD
	U[User]
	N[Noctis<br/>Lead agent / main user-facing contact]
	F[Active workflow step<br/>decides who acts next]
	I[Ignis<br/>Analysis / planning / review]
	G[Gladiolus<br/>Implementation / fixes / hardening]
	P[Prompto<br/>Recon / reporting / cleanup]

	U -->|request| N
	N --> F

	F --> N
	F --> I
	F --> G
	F --> P

Loading

• Noctis is the lead agent and the main user-facing contact.
• Ignis, Gladiolus, and Prompto are specialist teammates.
• The human user is referred to as User.
• Iris and Lunafreya are not part of the primary shipped browser workflow yet.

Quick Start

Prerequisites

• Node.js 20+
• npm 10+
• OpenCode CLI with an authenticated provider account

1. Initial setup

git clone https://github.com/atman-33/multi-agent-ff15.git
cd multi-agent-ff15
./first_setup.sh

first_setup.sh installs tmux and uv/uvx when needed, enables tmux mouse scrolling, creates local opencode.json from config/opencode.template.json, and initializes transport_mode: tmux-resident in config/settings.yaml when missing.

2. Authenticate OpenCode

source ~/.bashrc
opencode

Credentials are stored in ~/.opencode/.

3. Build once

./standby.sh --build

4. Start the dashboard

./standby.sh

The app runs at http://localhost:13000. When transport_mode is tmux-resident, interactive runs attach to the ff15 tmux session by default. Use ./standby.sh --no-attach to keep the current shell detached, or ./standby.sh --attach to force attachment explicitly.

Product Tour

Screen	Purpose
Noctis Team	Main mission surface for dispatch, activity, and mission history
OpenCode Sessions	Detailed session viewer for messages, tools, models, and execution context
Reports	Long-form report browser for reviewing outcomes after the live run
Projects	Project registry and workspace launch surface
Config	App settings and local config bootstrap helpers
MCP / Server	Runtime inspection for MCP configuration, health, and server state

Configuration

Tracked defaults live in config/.

• config/settings.yaml: language, shared_skills_root, transport_mode
• config/models.yaml: model assignments and switching metadata
• config/opencode.template.json: tracked template for the local opencode.json

opencode.json is user-local runtime config. It is bootstrapped by first_setup.sh and should not be treated as the canonical tracked template.

Use ./standby.sh --set-transport app-owned or ./standby.sh --set-transport tmux-resident to switch transport mode explicitly.

Development

Use the root npm scripts for web work:

npm run web:install
npm run web:dev
npm run web:build
npm run web:start
npm run web:typecheck
npm run web:lint
npm run web:test

npm run web:dev serves the app on port 5173 by default. ./standby.sh runs the production server on port 13000.

Repository Map

• web/: React Router app, loaders/actions, API routes, and UI state
• scripts/: setup, startup, transport, and config helpers
• config/: tracked settings, models, and local config templates
• projects/: project definitions used by the UI and workflows
• builtins/: localized built-in operation and instruction assets
• skills/: shared skills used by the workflow system
• openspec/: OpenSpec changes and specs
• docs/reports/: active report artifacts
• runtime/: generated runtime state and transport/session artifacts
• logs/: debug logs
• assets/: static images used by docs and the UI

Notes for Contributors and Agents

• Treat runtime/ and logs/ as generated state unless you are changing the code that produces them.
• Put active reports in docs/reports/.
• Prefer the root npm scripts for web/ tasks instead of running ad-hoc commands inside web/.

Credits

Inspired by multi-agent-shogun by @yohey-w and the original multi-agent-ff15 design.

License

MIT License.