What a AI Chatbot Development Platform actually does
Enables agencies to create, train, and deploy branded AI chatbots for each client with a shared multi-tenant platform, custom knowledge bases, and white-label client portals.
A white-label chatbot platform lets an agency create a unique chatbot for each client — trained on that client's website content, uploaded documents, or connected data sources — and deploy it on the client's domain with the agency's brand. The platform handles multi-tenant billing (Stripe Connect wallet), per-client knowledge-base management (RAG over uploaded docs), multi-channel deployment (web widget, WhatsApp, Facebook Messenger, Instagram, SMS), and human-agent handoff.
The market in mid-2026: Stammer.ai at $197/mo Agency tier is the reference product — it delivers genuine full white-label (your subdomain, your brand, Stripe Connect reseller billing) with BYO OpenAI key on the $497 Full SaaS tier. ChatLab at $360/mo is the main alternative. Botsonic, Chatling, and Botpress serve the mid-tier. The honest margin reality: your profit comes from token-cost arbitrage (BYO key = you control input/output costs vs Stammer's per-message markup at $197 tier) and the service layer on top — onboarding, customization, and ongoing management.
AI capabilities involved
Multi-tenant chatbot creation with per-client knowledge bases
RAG over documents, websites, and connected data sources
Multi-channel deployment (web, WhatsApp, Messenger, SMS)
Human-agent handoff with ticket creation
Who uses this
- Marketing and web agencies (5–100 clients) wanting to add AI chatbot SKUs to their service offering
- Vertical-niche resellers (real estate, dental, automotive) wanting a chatbot product tailored to their industry clients
- SaaS founders building a branded chatbot product without the engineering cost of building from scratch
- Digital marketing firms wanting to automate client lead capture and FAQ handling under their own brand
SaaS alternatives on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Stammer.ai
Marketing agencies (5–200 clients) wanting immediate branded chatbot revenue without engineering investment.
None
$197/mo (Agency — 20+ AI agents, unlimited end-clients)
$2,500+/mo (Enterprise — custom models, SLAs)
Pros
- +Full white-label: your subdomain (app.youragency.com), your logo, your colors — no Stammer branding.
- +Stripe Connect wallet billing lets you charge clients and set margins without a custom payment setup.
- +BYO OpenAI key on Full SaaS $497/mo = full token-cost control and no per-message markup.
Cons
- −Agency $197/mo uses Stammer's LLM pool — token economics are opaque below Full SaaS.
- −Healthcare clients need a custom Bedrock build — Stammer doesn't offer BAA at any tier.
- −Platform is OpenAI-centric — limited Claude or Anthropic model support.
- −White-label depth depends on plan; confirm all client-visible UI is de-branded before signing.
ChatLab
Agencies wanting a clean interface and willing to pay a premium over Stammer for perceived UX quality.
None
$360/mo (Agency — custom subdomain, white-label portal)
Pros
- +Full white-label with custom subdomain (app.youragency.com).
- +Clean UI praised by non-technical resellers.
- +Includes lead capture and CRM integration features.
Cons
- −More expensive than Stammer Agency ($360 vs $197) for similar feature depth.
- −No BYO LLM key option — you're dependent on ChatLab's API pricing.
- −Smaller ecosystem than Stammer — fewer channel integrations.
Botsonic / Writesonic
Individual businesses wanting to remove Writesonic branding from their own chatbot — not an agency resale platform.
Free with Writesonic plan
Brand removal as add-on (Plus $16.67/mo)
Pros
- +Cheapest branding-removal option in the market.
- +Good knowledge-base training from website URLs.
- +Integrates with Writesonic content generation tools.
Cons
- −Not full white-label — brand removal is an add-on, not a ribrand platform.
- −No client portal or multi-tenant management.
- −Built for individual use, not agency resale.
Botpress
Technical agencies building complex multi-step AI agents with custom integrations — not simple FAQ chatbot resale.
Free (limited)
$495/mo (Plus); Enterprise on quote
Pros
- +Most technical flexibility — custom LLM, custom actions, complex workflow logic.
- +Strong developer community and plugin ecosystem.
- +Visual flow builder for complex conversation architectures.
Cons
- −White-label requires Plus ($495/mo) or Enterprise — higher entry cost.
- −Complex for non-technical users — requires developer involvement to configure and maintain.
- −Plus tier white-label is partial — Enterprise required for full brand removal.
The AI stack
When building custom (post-500 clients or vertical compliance), the LLM stack choice determines both cost structure and compliance capability. For healthcare clients specifically, the LLM must be routed through AWS Bedrock or Azure OpenAI with a signed BAA.
Foundation LLM for chat generation
Powers the actual chatbot conversation — intent classification, response generation, context management
Claude Sonnet 4.6
$3 / $15 per M tokensPremium chatbot tiers — complex support conversations, nuanced FAQ handling
Claude Haiku 4.5
$1 / $5 per M tokensStandard chatbot tier — FAQ responses, lead capture, appointment booking prompts
GPT-5.4 mini
$0.75 / $4.50 per M tokensOpenAI-stack agencies wanting a cost-effective alternative to Sonnet
Our pick: Two-tier LLM routing: Haiku 4.5 for standard FAQ conversations (70% of volume), Sonnet 4.6 for complex flows (complaint escalation, multi-step service requests). This routing saves ~60% on LLM costs vs all-Sonnet.
Knowledge-base embeddings (RAG)
Embeds client knowledge-base documents for retrieval-augmented generation — the foundation of accurate, grounded chatbot responses
text-embedding-3-large
$0.13 / M tokensComplex knowledge bases with technical documentation, policies, or multi-domain content
text-embedding-3-small
$0.02 / M tokensStandard FAQ and product information knowledge bases
Our pick: text-embedding-3-small for standard client knowledge bases; text-embedding-3-large for clients with complex technical documentation or policy corpora. Offer as a plan tier differentiation.
Voice channel (optional add-on)
Enables voice-based chatbot interactions via phone or web RTC
Cartesia Sonic 3.5
~$35/M chars; <75ms TTFAAny real-time voice chatbot implementation where latency affects conversation quality
Our pick: Cartesia Sonic 3.5 for voice channels. Add voice as a premium tier add-on — not every client needs it, and it doubles the per-conversation cost.
Reference architecture
The custom chatbot platform architecture is fundamentally a multi-tenant RAG pipeline with per-client LLM context isolation. The hardest challenge is tenant isolation: each client's knowledge base must be completely isolated in Supabase pgvector, and per-client system prompts must be loaded per conversation without cross-tenant contamination.
Agency onboards new client and configures chatbot
Agency admin dashboardAgency creates client workspace. Configures: chatbot name, persona, brand colors, allowed topics, escalation keywords, system prompt prefix. Sets LLM tier (Haiku vs Sonnet). Creates API key for client's knowledge base.
Client uploads knowledge base documents
Client portal + Supabase StorageClient uploads PDFs, Word docs, URLs to crawl. Background job chunks documents into 500-word segments. Embeds each chunk via text-embedding-3-small (or large for premium). Stores vectors in pgvector with client_id tenant isolation. Supabase RLS ensures client can only access own vectors.
End-user sends message to chatbot on client's website
Embed widget + Supabase Edge FunctionWidget POSTs message to Edge Function with session_id and client_id. Edge Function: (1) retrieves client config (LLM tier, system prompt), (2) similarity-searches client's vectors for top 5 chunks, (3) calls configured LLM with context, (4) returns response. Cost: $0.005–$0.05 per turn.
Human handoff triggered by escalation keyword or user request
Edge Function + Supabase RealtimeIf escalation keyword detected or user clicks 'Talk to human', conversation tagged as handoff_requested. Agency dashboard shows real-time conversation for agent to take over. Optional: create Zendesk ticket via webhook.
Agency dashboard shows per-client analytics
Next.js admin + Supabase aggregationPer-client: message count, conversation count, handoff rate, top FAQ topics, monthly token usage. Client-level billing through Stripe Connect wallet.
Estimated cost per request
~$0.005 per FAQ-grade Haiku response with RAG; ~$0.02–$0.05 per nuanced Sonnet conversation turn. At 200 clients with 50 conversations/day each: ~$300–$1,500/month in LLM costs.
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Cost model for a custom-built platform at 200 end-clients, each generating 50 conversations/day.
Estimated monthly cost
$64.50
≈ $774 per year
Calculator notes
- LLM costs at 200 clients × 50 conversations/day × 30 days = 300K conversations/month: ~$13,500 in mixed Haiku/Sonnet costs. This assumes you're collecting $30–$100/client/month from agency markup — margin is tight without BYO key.
- With BYO key and LLM costs passed to clients per-usage: the platform's margin is entirely in the subscription fee and service layer.
- WhatsApp Business API: $0.005–$0.10 per conversation depending on message type and country — significant additional cost for high-volume WhatsApp deployments.
- Knowledge base re-embedding on document update: text-embedding-3-small at $0.02/M tokens; 1,000 chunks updated = $0.001. Negligible.
Build it yourself with vibe-coding tools
Buy Stammer Agency first ($197/mo). Only start a Lovable single-tenant prototype if you need to demonstrate a specific vertical feature (e.g. healthcare chatbot with specific compliance framing) that Stammer can't show. Multi-tenant + billing is 3–4 weeks of work beyond the prototype.
Time to MVP
1 weekend (single-tenant chatbot prototype); +3–4 weeks for multi-tenant + billing
Total cost to MVP
$25 Lovable Pro + $20 OpenAI credits = working single-client chatbot
You'll need
Starter prompt
Build a white-label AI chatbot prototype for a single client using Vite + React + TypeScript + Tailwind CSS with Supabase backend. This is a proof-of-concept, not the multi-tenant platform. Features: 1. Admin dashboard (auth required): upload knowledge base documents (PDF/TXT/URL). Each document chunked and embedded via OpenAI text-embedding-3-small. Stored in Supabase pgvector (knowledge_chunks table). 2. Chatbot config: set chatbot name, persona description, brand color, allowed topics, escalation keywords. 3. Chat widget preview: embedded chat interface (floating bottom-right button). Shows configured chatbot name and color. 4. Conversation flow: user message → Edge Function: embed message → similarity search knowledge_chunks top 5 → call Claude Haiku 4.5 with context → return response. Store in conversations/messages tables. 5. Escalation: if message contains escalation keywords OR user clicks 'Talk to human' → mark conversation as escalated → show admin notification badge. 6. Embed code: show <script> snippet for deploying the widget on any website. Database: knowledge_chunks (pgvector), chatbot_config, conversations, messages. Secrets: ANTHROPIC_API_KEY, OPENAI_API_KEY (for embeddings).
Paste this into Lovable
Follow-up prompts (run in order)
- 1
Add multi-tenant client management: add a 'clients' table. Each client has their own chatbot_config and isolated knowledge_chunks (client_id column + RLS policy). Admin can create/manage multiple client accounts. Each client gets a unique embed script with their client_id.
- 2
Add Stripe Connect wallet billing: integrate Stripe Connect so each client account has a connected Stripe account. Monthly usage billing (per conversation or flat rate) charged to client via Stripe Connect transfer. Admin sees revenue dashboard.
- 3
Add WhatsApp channel via Twilio: a Twilio WhatsApp webhook endpoint receives messages. Same Edge Function pipeline as web chat. Response sent back via Twilio API. WhatsApp phone number configurable per client account.
- 4
Add EU AI Act Art. 50 disclosure: all chatbot first messages must include: 'Hi! I'm [Name], an AI assistant for [Company]. How can I help you today?' — ensuring the AI nature is disclosed before the first substantive interaction.
Expected output
A working single-tenant chatbot trained on uploaded documents, deployed as an embeddable widget on any website, with admin dashboard for configuration and escalation monitoring.
Known gotchas
- !EU AI Act Art. 50 (August 2, 2026 enforcement deadline): every chatbot deployed to EU users must disclose it is AI before any substantive interaction. This is a platform-wide requirement — if your chatbot platform serves EU users, build disclosure into the first message template for all clients from day one.
- !Stammer's BYO key changes the margin math dramatically: on Stammer Full SaaS $497/mo with BYO OpenAI key, you control token costs and the platform fee becomes your fixed overhead. At 200 clients paying $50/mo markup, that's $10K MRR with $500/mo platform fee. Margin is strong. A custom build only wins when Stammer's feature limitations (no Anthropic models, no HIPAA, no vertical customization) cost you those 200 clients.
- !Per-tenant spend caps are critical: runaway agent loops (infinite retry loops, recursive function calls) can generate thousands of LLM calls per minute and thousands of dollars in API costs within hours. Build per-tenant monthly token caps with automatic cutoff before launch.
- !Supabase pgvector tenant isolation: without proper RLS policies, one client's knowledge base could be searched in another client's chatbot context. Test RLS isolation before client 2 goes live.
- !Meta app review for WhatsApp Business: getting WhatsApp Business API production access requires Meta business verification (1–2 weeks) and app review for your specific use case (1–4 more weeks). Do not promise clients WhatsApp before you have production access.
Compliance & risk reality check
Chatbot platforms face two non-negotiable compliance items in 2026: EU AI Act Art. 50 chatbot disclosure (August 2, 2026) and HIPAA BAA for any healthcare vertical clients. Both require platform-level implementation, not client-level.
EU AI Act Art. 50 — chatbot disclosure (Aug 2, 2026 deadline)
Effective August 2, 2026, any AI chatbot deployed to EU users must proactively disclose that the user is interacting with an AI system before the first substantive exchange. Failure to comply is an EU AI Act violation with fines up to 1.5% of global annual turnover. For a multi-tenant chatbot platform, this means platform-wide implementation.
Mitigation: Hardcode the AI disclosure into the first message template for all chatbots on your platform. The disclosure must happen proactively — not buried in terms of service. Example: 'Hi, I'm [Name], an AI assistant for [Company]. I'm here to help with your questions.' Train clients that they cannot remove this disclosure. If using Stammer, confirm their platform rollout plan for Art. 50 compliance.
HIPAA — healthcare vertical clients
Healthcare clients (clinics, dental practices, hospitals, pharmacies) who collect any patient-identifying information through the chatbot require a signed Business Associate Agreement (BAA) with the chatbot platform and its AI provider. Stammer and ChatLab default tiers do NOT provide BAA. Consumer-tier Claude, GPT, and other AI APIs are not HIPAA-eligible. Healthcare clients on a standard chatbot platform create HIPAA liability for both the client and the platform operator.
Mitigation: Do not onboard healthcare clients on standard chatbot SaaS without BAA coverage. For healthcare clients: build a separate deployment on AWS Bedrock (Claude available with BAA) or Azure OpenAI (HIPAA BAA available) with all required audit logging. Alternatively, explicitly exclude healthcare from your client acceptance criteria until you have a HIPAA-eligible infrastructure tier.
TCPA — SMS chatbot channels
SMS chatbot channels (Twilio SMS, WhatsApp) require documented prior express written consent per TCPA. If your chatbot platform enables clients to send automated SMS messages to their customers, your platform's terms of service must require clients to obtain proper TCPA consent before SMS activation.
Mitigation: Include TCPA compliance obligations in your platform terms of service. Require clients to confirm they have documented consent before activating SMS channels. Build a TCPA consent acknowledgment into the SMS channel activation flow.
Per-tenant spend caps — runaway agent costs
Chatbot platforms with agentic or multi-step capabilities can generate runaway LLM calls if an agent loop fails to terminate. A single misbehaving client chatbot can burn hundreds or thousands of dollars in API costs within hours. This is a platform-operator liability as well as a client cost issue.
Mitigation: Implement hard monthly token caps per client account with automatic conversation cutoff and admin notification when 80% of monthly cap is reached. Build circuit breakers that stop any conversation where a single session exceeds 50 LLM calls (indicating a likely loop). Log all sessions exceeding normal parameters for review.
Build vs buy: the real math
12–18 weeks
Custom build time
$25,000–$50,000
One-time investment
18–24 months (vs. Stammer Full SaaS at 500+ clients)
Breakeven vs buying
Stammer Full SaaS $497/mo with BYO OpenAI key: at 500 end-clients charging $50/mo markup average = $25K MRR - $497/mo platform fee = $24,503 net MRR. A $35K custom build would return this monthly run-rate after 1.5 months of operations — but only if you already have 500 clients (which takes time to build). The honest path: resell Stammer from day one, invest the margin in customer acquisition, and commission the custom build when you've validated 500 paying clients and identified specific features Stammer doesn't support.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship.
Discovery call (free)
30 minWe map your exact AI Chatbot Development Platform use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
12–18 weeksOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
12–18 weeks
Investment
$25,000–$50,000
vs SaaS
ROI in 18–24 months (vs. Stammer Full SaaS at 500+ clients)
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
What's the difference between Stammer Agency ($197/mo) and Full SaaS ($497/mo)?
The Agency tier at $197/mo uses Stammer's shared LLM pool — you don't control token costs, and Stammer's per-message economics apply. Full SaaS at $497/mo lets you bring your own OpenAI key (BYO key), which means your LLM costs are OpenAI's direct API pricing and you keep 100% of the margin between your client subscription price and OpenAI's rate. If you have more than 50 active clients generating significant chat volume, Full SaaS almost always pays for itself vs Agency.
How much does it cost to build a custom white-label chatbot platform?
A Lovable single-tenant prototype costs $25 + $20 API credits. A full multi-tenant platform with Stripe Connect billing from RapidDev runs $25K–$50K and ships in 12–18 weeks. Stammer Agency at $197/mo is the right answer for 95% of buyers — build custom only when you hit 500+ clients or need HIPAA-grade compliance that Stammer doesn't provide.
What is the EU AI Act Art. 50 deadline and how does it affect chatbot platforms?
August 2, 2026. From that date, all AI chatbots deployed to EU users must disclose that the user is interacting with AI before any substantive exchange. This applies to every chatbot on your platform. If you're on Stammer, confirm their platform has deployed this disclosure platform-wide. If you're building custom, hardcode the AI disclosure into the first message template for all chatbots and do not allow clients to remove it.
Can I use a white-label chatbot platform for healthcare clients?
Not on default Stammer/ChatLab tiers — they do not provide HIPAA BAA. Healthcare clients (clinics, dental, pharmacies) who collect any patient information through the chatbot require a BAA with both the platform and the underlying AI provider. For healthcare clients, build a separate deployment on AWS Bedrock (Claude + BAA) or Azure OpenAI (HIPAA BAA available) with full audit logging. Alternatively, explicitly exclude healthcare from your standard chatbot service offering.
How do I protect myself from runaway API costs from one misbehaving chatbot?
Implement per-tenant monthly token caps with automatic cutoff at 80% and 100% of the monthly limit. Build conversation-level circuit breakers that stop any conversation exceeding 50 LLM turns (a normal chatbot conversation is 3–10 turns; 50 indicates a loop). Notify yourself and the client when unusual usage patterns are detected. These safeguards are mandatory infrastructure for any multi-tenant chatbot platform.
Can RapidDev build a custom white-label chatbot platform for my agency?
Yes — RapidDev has shipped 600+ applications and can build a multi-tenant chatbot platform with per-client RAG knowledge bases, Stripe Connect billing, multi-channel deployment (web, WhatsApp, SMS), and EU AI Act Art. 50 compliance. Standard builds run $25K–$50K. Our honest recommendation: start with Stammer Agency ($197/mo) and build the client base to 200+ before commissioning a custom build. Book a free 30-minute consultation.
Want the production version?
- Delivered in 12–18 weeks
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.