Multi-Tenant White Label AI Assistants for Agencies

Agencies want to launch AI assistants across dozens of client sites without rebuilding infrastructure. Multi-tenant white label controls make that possible: each tenant gets isolated prompts, styling, quotas, analytics, and SLAs on top of a shared stack. Here is how to design it.

1. Tenant modeling

• Tenant record: Store domain allowlists, plan tier, quota caps, retention policies.
• Role-based access: Admin, Editor, Viewer roles per tenant; enforce via JWT claims or identity provider groups.
• Feature flags: Toggle premium features (private crawling, file uploads, advanced analytics) per plan.

2. Branding surface area

• Theme tokens for launcher button, header, conversation bubbles, fonts.
• Logo upload and powered-by toggle.
• Language overrides plus localized CTA copy for each embed.
• Custom welcome prompts per tenant or per embed.

3. Isolation controls

• Crawl jobs respect tenant-specific allowlists and robots rules.
• Retrieval filters include tenant_id and domain metadata to avoid data leakage.
• LLM usage quotas and billing counters roll up by tenant.
• Web analytics use embed_id + tenant_id to segment performance.

4. Operational guardrails

• Prompt versioning per tenant with audit logs.
• Crawl manifests stored with tenant tags, model version, embedding checksum.
• Google Chat alerts scoped to the tenant owner when thresholds trip.
• Retention policies (90 day chat logs default, configurable up to 730 days) stored on the tenant record.

5. Launch workflow for agencies

1. Discovery: Capture domain, sitemap, plan tier, branding assets.
2. Crawl: Run sitemap-first crawl capped by plan; log run_id for rollback.
3. Configure: Apply theme tokens, prompt persona, language defaults.
4. QA: Review per-embed metrics in staging, run hallucination checks, test SSO if required.
5. Deploy: Provide script snippet with tenant + embed tokens; monitor initial impressions and containment.

6. Reporting and billing

• Aggregate usage per tenant for invoices; include message counts, crawl minutes, LLM tokens.
• Provide agencies a dashboard showing every client’s impressions, opens, chats, containment, and feedback trends.
• Trigger billing grace alerts when a tenant nears quota; allow agencies to top up or adjust plan.

CrawlBot advantage

CrawlBot bakes these white label requirements into the core product: tenant isolation in MongoDB, per-embed analytics, adaptive thresholds, strict widget CSP, and plan-aware quotas. Agencies can focus on GTM and copy while the platform handles governance. If you are building your own stack, mimic these patterns to keep clients safe, on-brand, and measurable.