JARVIS Mission Control for DeepSeek-Harness

The open-source AI agent orchestration board, connected to DeepSeek-Harness.

JARVIS Mission Control is a Git-based command center for managing AI agents and human collaborators: a live Kanban board, task manager, activity log, messaging layer, and review workflow — backed by plain JSON files and a Node.js server. This edition connects it to DeepSeek-Harness (dsh), the plugin-kernel agent harness from DeepSeek: every dsh session appears as a task on your board, every turn and tool call streams into the activity log, and finished work lands in REVIEW for a human to approve.

Status: Experimental. dsh is a v0.1 developer preview with announced breaking changes. The bridge is validated against @deepseek-ai/dsh@0.1.0-rc.7 type definitions and covered by committed tests; event names are config-overridable so upstream renames are a config fix.

How It Connects

dsh treats everything as a plugin and broadcasts every model-visible moment on an append-only session event stream. Our bridge mounts inside the harness as a Cordis plugin and forwards that stream to Mission Control's REST API — no polling, no log scraping, no server changes:

┌────────────── dsh ──────────────┐        ┌──── Mission Control ────┐
│ agent loop → session event log  │        │ server (localhost:3000) │
│        │                        │  HTTP  │   POST /api/tasks       │
│        └─► mission-control ─────┼───────►│   PATCH /api/tasks/:id  │
│            plugin (this repo)   │        │   POST /api/logs/...    │
└─────────────────────────────────┘        │   → dashboard + WS      │
                                           └─────────────────────────┘

Quick Start

# 1. Fork and clone this repo
git clone https://github.com/YOUR-USERNAME/JARVIS-Mission-Control-DeepSeek.git
cd JARVIS-Mission-Control-DeepSeek

# 2. Start the Mission Control server + dashboard
cd server && npm install && npm start        # → http://localhost:3000

# 3. Install the bridge into the web profile. The dsh.bundle manifest
#    activates it automatically; no hand-written Cordis row is required.
npx @deepseek-ai/dsh@0.1.0-rc.7 plugin --profile web add dsh-plugin-mission-control

# 4. Run DeepSeek-Harness
npx @deepseek-ai/dsh web                     # → http://127.0.0.1:3080

Open the board at http://localhost:3000 and start a dsh session — the task appears as soon as the session does.

Full setup, config reference, and troubleshooting: integrations/deepseek-harness/README.md
Architecture analysis and design rationale: docs/deepseek-harness-integration.md

What's in the Box

The plugin ships with auth support (MC_AGENT_TOKEN / authToken → Authorization: Bearer), a serialized retry queue so a dead board never blocks the agent loop, and an awaited bounded flush() wired to dsh's session/flush durability checkpoint.

# Run the bridge's test suite (16 tests: rc.7 event fixtures + mock HTTP)
cd integrations/deepseek-harness/dsh-plugin-mission-control && npm test

MissionDeck Platform

This open-source repo is the engine. MissionDeck.ai is the platform around it — hosted dashboards (missiondeck.ai/mission-control/your-slug), visual agent building, and one-click deployment. A free API key from missiondeck.ai/auth connects your instance:

./scripts/connect-missiondeck.sh --api-key YOUR_KEY

Heritage & Roadmap

This project began as a port of JARVIS Mission Control (v2.1.0); the board, server, and data model carry over, and everything runtime-specific here targets DeepSeek-Harness only. See the repo issues for the roadmap:

1. Live end-to-end validation with a real DEEPSEEK_API_KEY
2. dsh-native cost tracking (per-step usage from assistant/message events → the dashboard cost cards)
3. Track new dsh preview releases with a compatibility canary and configurable event mappings

License

MIT — free to use, fork, and build on.