Relay: Open-Source Control Plane for OpenClaw AI Agents
What Relay Is
Relay is an open-source Electron desktop application that serves as a control plane for OpenClaw AI agents. If OpenClaw is the runtime, Relay functions as your local command center, providing the same workflow patterns as Claude Cowork but running on your own infrastructure with your choice of models.
Architecture and Components
The system follows a clear separation between control and execution planes:
- Relay (Control Plane): Desktop app where you define tasks, review deliverables, and approve actions
- OpenClaw Gateway (Execution Plane): Runtime that runs on your infrastructure (local, VPS, or custom URL) and executes tasks
Communication between Relay and OpenClaw happens via WebSocket/API connections.
Key Features from Source
- Dispatch & Chat: Give tasks in natural language, agent plans steps, review polished deliverables
- Governance: Approval gates for file operations, shell commands, and data sends; exportable audit trail; cost tracking per task
- Configuration: Schedule recurring tasks, browse and edit agent memory, manage connectors (Slack, Notion, GitHub, Jira, etc.), set project working folders
- Model Flexibility: Routes tasks to any LLM backend including Claude, GPT-4, Llama, Gemini, Mixtral, or custom endpoints
- Data Sovereignty: Processes files on your infrastructure rather than Anthropic's servers
Example Workflow
The source provides a concrete example of a scheduled daily briefing:
- In Relay: Create schedule for "Daily 8am", set connectors (Slack + Notion)
- OpenClaw executes: Cron fires at 8am, agent reads project files, pulls Slack threads and Notion pages, calls your chosen LLM, writes briefing to memory
- Briefing appears in Relay for review, approval, or redirection
Target Use Cases
Relay addresses three structural limitations of Claude Cowork: data sovereignty concerns for regulated industries, model lock-in (only Claude), and compliance gaps where activities aren't captured in standard audit logs. This makes it particularly relevant for organizations with strict data policies or those needing to use multiple LLM providers.
📖 Read the full source: HN AI Agents
👀 See Also
engram v3.4.0 Adds Anthropic Plugin to Keep Claude Code Running Under New Rate Limits
engram v3.4.0 introduces a dedicated Anthropic plugin for Claude Code, adding three skills to manage costs, query context, and surface errors. Install with `/plugin install engram` or `npm install -g engramx@latest`.
Hypura: Storage-tier-aware LLM inference scheduler for Apple Silicon
Hypura is a Rust-based inference scheduler that places model tensors across GPU, RAM, and NVMe tiers to run models exceeding physical memory on Apple Silicon Macs. It enables running a 31GB Mixtral 8x7B on a 32GB Mac Mini at 2.2 tok/s and a 40GB Llama 70B at 0.3 tok/s where vanilla llama.cpp crashes.
TeamOut AI Agent for Company Retreat Planning
TeamOut has launched an AI agent that plans company events through conversation, handling venue sourcing, vendor coordination, flight cost estimation, itinerary building, and project management. The system uses multiple LLMs and specialized tools to manage planning as a stateful coordination problem.
Developer Builds GALA Programming Language with Claude Code, Notes Strong Typing Improves AI-Generated Code Reliability
A developer built GALA, a functional programming language that transpiles to Go, using Claude Code extensively. The language features sealed types, exhaustive pattern matching, immutability by default, and monads, with Claude helping implement type inference, pattern-matching transformers, and fixing over 40 bugs.