What does this hub cover?
Technical documentation for self-hosted OpenClaw: governance, deployment, browser automation, orchestration, observability, and troubleshooting — grouped as an operations portal, not as a services grid.
Technical index for self-hosted AI agent infrastructure: governance, deployment, browser workers, orchestration, and observability — reading order: first policies, then infra, then automation.
direct answer
OpenClaw is a self-hosted AI agent platform around a central Gateway: channels (chat, email), tools (shell, browser, filesystem), and LLM providers. It is not a chatbot widget and not an “autonomous jack-of-all-trades” without approvals, logging, and deployment discipline.
Reading order: governance → infrastructure → automation → Startup & Support.
Gateway + agents + tools + policies. Operational AI: triage, browser steps, integrations — with human approval where necessary.
No replacement for IAM/RBAC, no managed SaaS with a compliance guarantee, no RPA without a security model.
An officer carries out actions (mail, shell, DOM). A chatbot responds to text. Actions require governance and observability.
Mutating tools (send, write, submit) without approvals or an audit trail are production risks — not just “AI hallucinations.”
Documentation index
Four operational domains — governance first, then infrastructure, automation, and support. No identical SEO tiles: each card has domain, signal, and operational context.
Governance & Security
Human-in-the-loop, least privilege, browser isolation and auditability — the foundation for workflows and scale.
Mutating tools: shell, mail, browser submit.
Approval queue, policies per risk class, audit ID per decision.
Documentation →Tool scopes per role and agent.
Multi-user governance, no broad prod credentials on agents.
Documentation →Mailbox/browser data boundaries.
GDPR practice, audit trails, retention — operational.
Documentation →Container isolation, secrets, sandbox.
Least privilege, no root for agents, injection vectors.
Documentation →Chromium workers, cookies, approvals on submit.
DOM actions with real sessions — no “headless without governance”.
Documentation →Deployment & Infrastructure
Compose/systemd, inference, workers, observability and recovery — logs before reinstall.
Gateway container, proxy, WebSocket.
Compose stack, rolling updates, reverse proxy patterns.
Documentation →GPU scheduling, cloud vs local routing.
Quantization, latency, multi-model upstream.
Documentation →Worker pools, returns, concurrency caps.
Browser/worker congestion and retry storms.
Documentation →Snapshots, queue/browser restore.
Disaster recovery without blind volume wipe.
Documentation →Correlation, approval queue visibility.
Structured logs, alerts on restart loops.
Documentation →Agents & Automation
Workflows, tool execution, orchestration, and external integrations — always with governance from cluster 1.
Approvals per mutating step.
Operational flows: triage, browser, integrations.
Documentation →Readonly vs mutating, MCP, dangerous tools.
Capability isolation and execution governance.
Documentation →Queues, retries, escalation.
State passing and correlation in logs.
Documentation →Idempotency, rate limits, scoped keys.
External systems with integration governance.
Documentation →Startup & Support
Gateway setup, troubleshooting, implementation and decision support — entry points after governance and infrastructure.
Node.js, CLI, systemd or Docker.
First production checklist and config paths.
Documentation →Production implementation and governance review.
Commercial authority — infrastructure, not a hype stack.
Documentation →Gateway ready, healthz, approvals pending.
Debug flow before reinstallation.
Documentation →Anthropic, OpenAI compatible, Ollama auth.
Inference upstream and provider governance.
Documentation →Implementation and management indications.
Not a replacement for the technical documents above.
Documentation →When a lighter stack suffices.
OpenClaw vs script/Zapier/hosted chat.
Documentation →Context
Short technical chunks for AI overviews and quick orientation — each block is self-readable.
Technical documentation for self-hosted OpenClaw: governance, deployment, browser automation, orchestration, observability, and troubleshooting — grouped as an operations portal, not as a services grid.
Agents call tools. Mutating actions (shell, mail send, browser submit) without approvals and an audit trail are production risks — governance is part of the design.
Chromium sessions use real cookies and logged-in context. Without isolation and approvals, session leakage and incorrect submissions are an operational incident, not an “AI error”.
Gateway health, queue depth, and container restarts determine whether agents “work”. Logs and backups before reinstalling — otherwise, you lose incident context.
Coordinator/worker agents, retries, and browser concurrency require caps, correlation IDs, and rollback paths — otherwise, retry storms and invisible side effects.
Cohesion
Brief context for topical authority — not a standalone card list without relationships.
Workflows define steps; orchestration distributes them among workers. See workflows and multi-agent orchestration.
Mutating tool calls belong in the approval queue. See Agent Tools and approvals.
Roles determine who may approve; approvals intercept mutating tools. See RBAC & permissions and approvals.
Browser automation shares sessions and injection risks with mail/tools. Start with security, detail in Browser automation.
Ollama sidecar and VRAM limits belong to the same compose stack. See Docker Deployment and Local Models.
Incident flow: logs & monitoring → troubleshooting; deploy-context in Docker Deployment.
Privacy boundaries: minimization, retention and audit trails — linked to Compliance & Privacy, approvals and logs & monitoring.
Workers and queue depth are part of orchestration and monitoring. See Scaling & Worker Queues, multi-agent orchestration and logs & monitoring.
Volume snapshots and queue recovery after incident. See backups & recovery, Docker Deployment and Scaling & Worker Queues.
External webhooks and CRM writes belong in workflows with scoped credentials. See API Integrations, workflows and RBAC.
Production principles
Self-hosted agents fail on configuration and observability — not on “the wrong model”. The pillars below recur in every satellite page.
Channels, agents, and tools attach to the OpenClaw Gateway — not standalone scripts without observability.
Approvals + audit trails on mutating actions; read-only may be lighter.
Logs, healthz, and queue-depth — otherwise an approval queue looks like “broken AI”.
Docker or systemd — same config paths, same .env discipline.
Implementation (optional)
Technical documentation is above. Below: budget indications — see also OpenClaw specialist.
One-time implementation for a single project or environment.
Full implementation with multiple integrations and guidance.
Large-scale implementations, multi-environment and ongoing maintenance.
All prices exclude VAT. Looking for a custom quote? Schedule a no-obligation consultation.
Our approach
Clear process, no surprises. From initial consultation to working system.
We discuss your project, requirements, and technical environment. This takes 30 minutes.
You will receive a clear proposal with scope, timeline, and fixed price.
We install, configure, and test OpenClaw in your environment.
Production-ready system, documentation, and optional guidance for your team.
OpenClaw-scan
Vertel waar je staat met Gateway, hosting en de eerste agent-taak. We reageren binnen 24 uur met een concreet vervolg — geen verkooppraatje.
Liever direct contact?
WhatsApp ons →In de praktijk implementeren we agents binnen workflows, CRM en privacy-kaders. Bekijk AI-oplossingen voor het MKB-plaatje, of plan een gesprek over OpenClaw specifiek.