Guides
Self-hosted customer service software: why and how to choose
Your customer-service system holds every client conversation, contact and sales lead. Running it self-hosted — on your own servers or dedicated cloud — is increasingly common in finance, healthcare, government and compliance-heavy industries. This guide covers when self-hosting pays off, how it really differs from SaaS, and a concrete path to production.
Why self-host customer service
Three drivers dominate. Compliance: regulators, security assessments or a hard “data never leaves our DC” rule take SaaS off the table. Cost: per-seat subscriptions compound — over three to five years a growing team often pays multiples of a one-time self-hosted license. Integration: support must plug deep into your own ticketing, CRM and knowledge bases, and self-hosting gives source-level freedom.
Conversely, a small team wanting the fastest launch with zero data-retention requirements is still better served by fully managed SaaS — decide which camp you are in first.
| Dimension | SaaS support | Self-hosted support |
|---|---|---|
| Data storage | Vendor cloud | Your servers; can run fully air-gapped |
| Cost model | Recurring per-seat fees | Mostly one-time licensing (free tiers exist) |
| Integration depth | Open-platform APIs within vendor limits | Server APIs + source-level client customization |
| Time to launch | Fastest, sign up and go | Needs a server and basic ops |
| Xinchuang (domestic stack) | Rarely applicable | Adaptable to domestic CPUs and OSes |
Four hurdles to clear when self-hosting
Self-hosting is more than installing software. In order of deployment: topology (single node or cluster), reliability (no lost messages, no dropped agent sessions), AI quality (knowledge-base answers decide how many heads you save), and day-two ops — upgrades and backups.
- Deployment: one compose command brings up DB + Redis + the server; a single node carries tens of thousands of concurrent connections
- AI: RAG knowledge base (pgvector) + function calling with your own model (OpenAI-compatible / DeepSeek / Dify / Coze)
- Human agents: web widget, chat window and email / WhatsApp / Telegram / Messenger / Instagram / WeChat channels land in one agent workspace, with instant human↔AI handover
- Ops: migrations run with upgrades automatically; database backup is one command daily
How IMCore does self-hosted support
IMCore builds support on top of an IM engine: AI agents (RAG + tool calls) and human agents share one messaging plane, a web widget embeds into your site, an email channel works out of the box, and one agent workspace brings it together. The Basic edition is free — run it on your own server before committing; deployment is Docker Compose or Kubernetes, with Xinchuang environments supported.
For saturation-test numbers (ten thousand concurrent connections on one 4C8G node) and the clustered topology, see “Self-hosting a chat server”.
FAQ
How does the cost of self-hosted support work?
Mostly one-time licensing plus your server and ops; no recurring per-seat fees. The more agents and the longer the horizon, the bigger the amortized advantage over subscriptions. IMCore’s Basic edition is free; commercial editions offer yearly or buyout terms.
How much ops effort does self-hosting take?
The single-node form is one Docker Compose command; day-to-day work is mostly upgrades and database backups (one command each). Clustering without client changes starts with the Pro edition. Small teams without dedicated ops can sustain it.
How does AI support run in a self-hosted environment?
The knowledge base (RAG) runs in your own PostgreSQL (pgvector extension); the model side plugs into any OpenAI-compatible service, DeepSeek, or Dify / Coze / custom workflows — you control both the data and the model calls.
Keep reading
Head-to-head comparisons
/en/compare/ →Production deployment with Docker Compose
/en/docs/deploy-docker-compose/ →Self-hosting a chat server
/en/docs/self-hosted-chat-server/ →Self-hosted live chat: a practical guide
/en/topics/self-hosted-live-chat/ →Self-hosted AI customer service: an on-prem selection guide
/en/topics/self-hosted-ai-customer-service/ →Air-gapped instant messaging: selection and deployment
/en/topics/air-gapped-instant-messaging/ →READY TO OWN THE CONVERSATION?
Bring your workload. We’ll size the architecture.
Share your concurrent connections, daily message volume and deployment environment for a concrete edition and topology recommendation.
We usually reply within one business day