Inspired by OpenClaw

The inspiration

OpenClaw is a published open-source pattern for local-first autonomous AI agents — a long-running agent that monitors a set of conditions, reaches out when something needs attention, and accepts instructions through the messaging channels its operator already uses. The pattern demonstrated several ideas that shaped SydClaw's product surface: proactive monitoring instead of pull-based prompting, a unified session that follows the operator across channels, and high-friction destructive actions that route through the operator's messaging app for explicit approval.

OpenClaw is what an individual technical user might run for themselves. SydClaw is what a regulated business needs from the same idea: the same proactive, channel-aware, approval-gated behaviour, but inside an audited, multi-tenant, compliance-first managed deployment.

What we adopted

The heartbeat pattern. A scheduled job — every 30 minutes per organisation in our case — checks a configurable list of conditions: unanswered emails older than four hours, CRM deals with no activity in seven days, project deadlines approaching, audit-prep evidence missing for a Practice Standard indicator. If nothing needs attention, the heartbeat suppresses notification entirely. The principle is the same as OpenClaw's: monitoring should be proactive, but notifications should be earned.

Multi-channel session unification. An operator who starts a thread on Slack and continues on email should be talking to the same agent with the same context. SydClaw routes inbound messages from Gmail, Microsoft 365 Outlook, Slack, Teams, and the in-app chat into a unified session store keyed by the user's organisational identity, not by channel. The agent doesn't restart its memory because the operator switched apps.

Approval-via-message for high-risk actions. When the agent wants to take a destructive or high-cost action (sending external email, processing an invoice over a threshold, lodging a Commission notification), the request is forwarded to the operator's messaging surface. The operator approves with a short acknowledgement; the agent resumes. This is OpenClaw's approval-via-message pattern, applied to enterprise workflows.

What we built differently

Multi-tenant security at the database layer. OpenClaw runs locally for a single user. SydClaw runs as dedicated infrastructure per client, with row-level security at the PostgreSQL layer enforcing organisation isolation on every query. A stolen API key from one tenant cannot read another tenant's data — the database itself enforces the boundary, not the application code.

Privacy router and PII tokenisation. OpenClaw is built for an individual user who is comfortable with their own data going to whichever LLM provider they configure. SydClaw operates in regulated industries where client PII cannot leave the platform in plaintext. We added a privacy router that tokenises 17 categories of personal information before any external model call. The model never sees raw client data; reverse-mapping happens only inside the client's own infrastructure.

Hash-chained audit trail for the regulator. OpenClaw's logging is operationally sufficient for an individual. Australian professional services firms — accountants, lawyers, brokers, NDIS providers, aged care providers — operate under regulatory frameworks that require explainability of automated decisions. SydClaw's audit log uses a SHA-256 hash chain so that any tampering after the fact is mathematically detectable, with RESTRICTIVE row-level policies preventing UPDATE or DELETE on the audit table.

Vertical compliance modules. The OpenClaw pattern is general-purpose. SydClaw bundles vertical-specific modules — NDIS Practice Standards mapping, Aged Care Quality Standards readiness, AFSL Best Interests Duty screening, TASA 2009 audit trail, APRA CPS 234 controls — that would be out of scope for a generic personal-agent project.

Managed deployment, not self-host. OpenClaw is a project a developer runs locally. SydClaw is a managed service: AI Lab Australia provisions the dedicated cloud instance, configures the integrations, authors the policies, and maintains the deployment. The professional services firms we sell to do not have an in-house team to run an open-source agent framework — and the regulatory exposure of self-hosting an unaudited AI system is a non-starter for them.

Credit and licence

SydClaw does not embed, fork, or redistribute the OpenClaw codebase. The inspiration is conceptual — the heartbeat pattern, multi-channel session unification, and approval-via-message — implemented independently in TypeScript on top of Inngest, Next.js, and Supabase. Nothing in SydClaw's runtime depends on the OpenClaw project.

We credit OpenClaw because honest attribution to the open-source ideas we built on is the right thing to do, and because being clear about lineage helps prospective clients understand what part of SydClaw is novel (the compliance, privacy, and managed-service layers) and what part is a careful adaptation of an established pattern (the agent-loop and notification design).