Self-hosted AI customer service is support software that runs on infrastructure you control, answering customers with an AI model billed through your own API keys. Conversations, customer records and your knowledge base never leave your servers — and you buy the software once, instead of renting it per resolution, per seat, per month.
A self-hosted deployment has four parts, all running under your control: a chat widget your customers talk to (web and WhatsApp), the answer engine that grounds every reply in your own knowledge base and live data (orders, deliveries, account status), an admin console where your team watches the queue, approves sensitive actions and takes handoffs, and your model API key — the single external call, made through your own OpenAI or Anthropic account. Nothing phones home to the vendor; there is no vendor in the data path.
| Self-hosted (Clare AI) | SaaS (Intercom, Zendesk, Ada…) | |
|---|---|---|
| Where customer data lives | Your servers | Vendor’s cloud + their subprocessors |
| Pricing model | One-time license + ~1¢/conversation model usage | Per resolution ($0.49–$2.00) and/or per seat, forever |
| Cost as you grow | Flat — volume barely moves the bill | Grows with every conversation |
| If the vendor disappears | Keeps running; source code included on higher tiers | Product dies with the vendor |
| Restricted industries (e.g. iGaming) | Your server, your rules | Many vendors restrict or refuse |
| Maintenance | Yours (or an optional care plan) | Vendor handles it |
| Compliance posture | Assessed under your own controls — details | Vendor questionnaires, DPAs, certifications |
Three numbers: a one-time license (Clare AI: $1,999–$10,999 depending on how much is done for you), model usage at roughly a cent per conversation billed by your provider, and hosting — an ordinary VPS, which Starter and Professional tiers include for the first year. At 5,000 conversations a month that totals about $3,799 over three years; the same volume on per-resolution pricing runs $88,000–$360,000 at 2026 list rates. Full price breakdown: how much AI customer service costs in 2026.
· Your support volume is high enough that per-resolution billing hurts
· Customer data must stay under your control (finance, health, iGaming)
· Mainstream SaaS restricts your industry
· You want software that survives any vendor
· You have — or can rent — basic server admin (or take the care plan)
· Your volume is tiny and per-resolution bills are genuinely small
· You want zero responsibility for any infrastructure, ever
· You need a deep app-marketplace ecosystem more than data control
· Nobody on your side can own an admin password
Delivery within a week of payment: the software on your server, knowledge base loaded, widget ready to embed, admin console handed over in writing. Professional and Enterprise include the full source code and deployment scripts. Every purchase carries a 30-day deployment guarantee — live and answering your customers, or your money back. See pricing, the resolution-rate methodology, and how we compare to Intercom, Zendesk and Salesforce.
The chat bubble on our homepage is the actual self-hosted product, answering from our own knowledge file. That’s the demo.
Structurally, it is the most private option available: conversations, customer records and your knowledge base stay on servers you control, and the only external call is to your own AI-model provider under your own account. Security then depends on your server hygiene — HTTPS, key storage, access control and backups.
You bring your own API key from OpenAI or Anthropic (a five-minute signup). The software calls the model through your account, so usage — roughly one cent per conversation — is billed to you directly at cost, with no markup.
An ordinary cloud VPS is enough for most businesses — the AI heavy lifting happens at the model provider, so the server mostly routes conversations and stores data. Clare AI Starter and Professional tiers include a configured cloud server with the first year of hosting.
Yes — escalation is a core feature, not an afterthought. The AI hands hard or sensitive cases to your team with full conversation context, and a human-approval step can gate sensitive actions like refunds.
At 5,000 AI-handled conversations a month over three years: Intercom Fin at $0.99/resolution is about $178,200 and Zendesk at $1.50 is about $270,000, at public list rates before seats and platform fees. A self-hosted Clare AI license plus model usage is about $3,799 all-in over the same period. Run your own volume.