AgentDeck is a local-first workspace for multi-agent coding and low-code orchestration.
The project is in early development. Current work is focused on turning the product design into a small, testable monorepo: runtime detection, local daemon, agent registry, chat workspace, workflow core, workflow canvas, and safe patch approval.
AgentDeck is designed to combine:
- Slack-like multi-agent collaboration in channels and threads.
- Local runtime detection for Codex, Claude, Ollama, LM Studio, Node, Git, and related tools.
- Low-code Agent and Workflow DAG orchestration.
- Safe chat-to-workflow editing through patch proposals, validation, diff preview, approval, and rollback.
The MVP focuses on:
- Runtime Detector
- Agent Registry
- Chat Workspace
- Local Daemon
- Workflow Canvas
- Workflow Patch Approval
The MVP intentionally excludes multi-tenant enterprise permissions, cloud-hosted execution, billing, plugin marketplace, full RAG, custom model training, and full Dify/Coze-style app publishing.
- Product design:
AGENTDECK_PRODUCT_DESIGN.md - Original project brief:
agentdeck-project-brief.md - Development plan:
docs/development/README.md - Task backlog:
docs/development/task-backlog.md - Architecture:
docs/architecture.md - Runtime adapters:
docs/runtime-adapters.md - Workflow schema:
docs/workflow-schema.md - Security model:
docs/security-model.md - Contributing guide:
docs/contributing.md - Agent instructions:
AGENTS.md - Claude Code instructions:
CLAUDE.md
AgentDeck is open source under the MIT License.