Products, Stack, and Vocabulary

Products and offerings

• OpenClaw: AI agent platform for orchestrated, use-case-driven workflows.
• Automations: workflow systems for repetitive operational tasks with policy guardrails.
• Websites: high-conversion sites with analytics, content operations, and SEO/AEO foundations.
• Apps: web applications for internal workflows, customer portals, or productized services.
• AI-powered features: retrieval, triage, summarization, copilots, and workflow assistants inside products.

Model and LLM strategy

Use model variation by task, not brand hype. Different tasks need different tradeoffs in latency, cost, reasoning depth, and reliability.

• Fast models: classification, extraction, triage, and lightweight routing.
• Reasoning models: planning, architecture decisions, and multi-step synthesis.
• Hybrid routing: dispatch between model classes based on confidence and risk level.
• Fallbacks: failover paths for outage resilience and predictable service behavior.

Key stack items

• Orchestration loop (heartbeat): decide, act, evaluate, repeat.
• Agent roles: planner, worker, reviewer, evaluator, human approver.
• Tooling layer: permissioned actions for code, data, APIs, and external services.
• MCP integrations: standardized interfaces for external tools and resources.
• Observability: logs, traces, outcomes, and error analysis.
• Safety controls: policy checks, approvals, rollback paths, and auditability.

Common vocabulary

• Heartbeat: runtime control loop for orchestration.
• Soul: persistent identity, memory, and operating constraints.
• Agent: role-specific reasoning unit in a workflow.
• Tool: executable capability an agent can call.
• MCP: protocol for accessing tools/resources with consistent interfaces.
• Human-in-the-loop: required person approval on high-impact actions.
• Eval loop: recurring measurement process to prevent regressions.