Command interface // Online

MECHBAY

A BattleTech-inspired command bay for AI coding agents.

Five mechs. Real agents. Your projects are the battlefield.

▸ VIEW ON GITHUB ▸ TRY THE DEMO (NO API KEYS)

Animated demonstration of a mech deploying to a project facility, streaming agent output, and returning with a mission debrief — Bay feed // Deployment loopLive telemetry

Mission protocol

From hangar to git diff.

The isometric bay is the interface. Every deployment crosses the line into a real project, a real process, and a reviewable result.

STEP 01

DEPLOY

Drag a mech onto a facility: a real project directory represented in the bay.

STEP 02

EXECUTE

A real agent process launches in that directory while live output streams into the HUD.

STEP 03

DEBRIEF

The mech returns with a Mission Debrief, a real git diff, and new mission history written to memory.

Active chassis // 05

Choose your mech.

Each chassis has a specialty, a real runtime, and a persistent identity that travels from mission to mission.

Heavy assault

Atlas-Prime

RUNTIME // Claude Code

Built to carry the heaviest missions and keep pushing when the codebase pushes back.

Surgical strike

Marauder-Prime

RUNTIME // Codex

Precise, direct, and happiest when a difficult change needs a clean line through the noise.

Recon scout

Raven-Prime

RUNTIME // Kimi

Scans first, reports clearly, and streams ▸ INTENT / ◆ FINDINGS thought cards as it works.

Ranged multimodal

Catapult-Prime

RUNTIME // Gemini

Brings long-range context and multimodal firepower to missions that extend beyond text.

Swarm courier

Locust-Prime

RUNTIME // Bring your own agent

Fast, adaptable, and ready to carry any CLI you bring through the same deployment loop.

Persistent context

The soul layer.

These are not disposable chat tabs. Every mech carries a soul.md persona and a memory.md mission history into each deployment. The context compounds. The roster remembers what it tried, what worked, and what to check next time.

soul.mdmemory.md

Atlas-Prime / memory.md

01 ## Mission: Reactor Control
02 - Facility: reactor-control
03 - Objective: survey + calibrate telemetry
04 - Edits: src/telemetry.ts, mission-report.md
05 - Result: clean diff, debrief filed
06 - Next time: check sensor thresholds first

System schematic

Under the hood.

A desktop architecture designed to keep cinematic presentation separate from process execution, with every boundary inspectable.

MechBay desktop app showing the isometric bay, mech roster, facilities, and command HUD — Renderer capture // Isometric bay

Desktop shell: Electron main, preload, and renderer processes with a Phaser isometric scene.
Runner boundary: One interface across five agent runtimes, including bring-your-own CLI.
IPC contracts: A typed registry governs messages crossing Electron process boundaries.
Key custody: API credentials encrypted at rest through the operating system’s safeStorage.
Mission control: Three concurrent deployments, a FIFO queue, and crash recovery.
Verification: 297 unit and integration tests. Read-only facility browsing. Real git diffs.
Simulation: A built-in demo runtime exercises the full loop without API keys.
License: Open source under MIT. Inspect it, extend it, bring your own agent.

Zero-key launch

Enter the bay.

Clone the repo and run the complete simulated deployment loop. No provider account, API key, or agent CLI required.

PowerShell / Terminal

git clone https://github.com/samalbanese/mechbay.git
cd mechbay
npm install
npm run demo

Have a real agent CLI? Launch the development bay with npm run dev.