Comparison2026-07-27·9 min read·By Nedas Višniauskas

Best Open-Source LLM Customer Support Platforms (2026)

"LLM-first" and "self-hostable" used to be a rare combination in customer support tooling. In 2026 there's a real field of open-source options - but most of them share the same architecture: a web widget or help-center overlay, embedded via script tag or WebView. If your product is a native iOS or Android app, none of that reaches the place where users actually get stuck.

This is a practical roundup of the strongest open-source, LLM-first support platforms available today, including where each one fits and where the category still has a gap.

Open-Source LLM Customer Support Platforms Compared

|---|---|---|---|---|

Tiledesk

Tiledesk is the most complete of the general-purpose open-source options: RAG knowledge bases, LLM workflows, Ollama/vLLM support, a shared human-agent inbox, and channels spanning web chat, WhatsApp, Telegram, and email. It's a genuine self-hosted alternative to a SaaS help desk with AI layered in, and its human-in-the-loop handoff carries conversation context to the agent taking over.

The tradeoff is operational surface area - MongoDB, Redis, messaging infrastructure, and vector storage all need to be run and maintained.

Chatwoot

Chatwoot is the most mature project on this list: a polished shared inbox, live-chat widget, email support, and a large open-source community. Its Captain AI system covers customer-facing automated answers, agent copilot drafting, and knowledge-base suggestions - but the fully self-hosted version of Captain is documented as an Enterprise Edition feature, so the unrestricted community install is closer to "support-first with AI" than "LLM-first."

ChatterMate

ChatterMate is purpose-built as an AI support product rather than a general chatbot framework - RAG-based answers, human handoff, visual workflows, and integrations with Shopify, Slack, and Jira. The Apache 2.0 license is the most permissive on this list if you plan to modify or commercialize the implementation. It's newer, so the operational track record is shorter than Tiledesk's or Chatwoot's.

TGO

TGO is built around teams of specialized agents rather than a single assistant - billing, technical support, and routing can each be their own agent, coordinated through MCP tools and a shared knowledge base. It ships a Flutter widget for mobile, which is closer to "in-app" than most of the list, though it's still a widget embedded in a webview rather than a native SDK.

AnythingLLM

AnythingLLM isn't a help desk - it's the fastest way to stand up a private RAG assistant and drop it on a website. If most of your support volume is documentation-answerable, it's the lowest-effort option here. Ticket management, SLA tracking, and robust agent handoff aren't its job; teams typically pair it with a separate support system for anything beyond Q&A.

Dify

Dify is a visual builder for RAG pipelines, chatflows, and tool-calling agents, with embeddable web chat and OpenAI-compatible model support. It's the right tool when you have engineers who want to design custom support logic rather than adopt someone else's product opinions - but it doesn't come with a ticketing inbox or human-agent workspace out of the box, and its license includes restrictions on operating multi-tenant services that are worth reading closely.

ResolveKit

Every product above embeds through a web widget, webview, or Flutter shell layered on top of a mobile app. ResolveKit takes a different approach: it's a native iOS (Swift) and Android (Kotlin) SDK that puts the AI agent directly inside your app's own UI, with access to real app state, screen context, and product workflows - not just what a widget can scrape from the DOM.

It's also the only project on this list that can take approved action inside the product, not just answer questions. Tool calls run through operator-defined approval policies - autonomous, approval-required, or disabled - and every session produces a full trace: context, proposed action, approval decision, and outcome. The backend is AGPL-3.0 and the SDKs are MIT; self-hosting is free, and you bring your own LLM via any OpenAI-compatible provider (OpenAI, Anthropic, Gemini, or local models through Ollama). There's no per-resolution platform fee - teams typically land around $0.05 per resolution on Gemini Flash-Lite, all LLM cost, no ResolveKit markup.

Which Should You Choose

General web-based help desk with AI added: Chatwoot, if the mature inbox and community matter as much as the AI layer.
Full LLM-first platform with human escalation: Tiledesk, for the most complete out-of-the-box combination.
Documentation-only support bot: AnythingLLM, for the lowest-effort RAG widget.
Custom-built support agent: Dify, if you have engineers who want to design the logic themselves.
Multi-agent support architecture: TGO, if billing, technical, and routing agents should be separate.
Permissively licensed, newer AI-native product: ChatterMate, for Apache 2.0 flexibility.
Native mobile app with real support volume: ResolveKit - the only option here that's a native SDK rather than a web widget, and the only one that can execute approved actions instead of only answering questions.

Frequently Asked Questions

Is there an open-source LLM support tool for native mobile apps?

Yes - ResolveKit. It ships as a native iOS (Swift) and Android (Kotlin) SDK rather than a web widget or WebView, so the AI agent has access to real app state and screen context. The backend is AGPL-3.0 and the SDKs are MIT-licensed; self-hosting is free with bring-your-own LLM keys.

What's the best open-source alternative to Intercom Fin?

Among the options here, ResolveKit is the closest architectural alternative for mobile apps - it can execute approved actions with operator guardrails, not just answer questions, and costs roughly $0.05 per resolution with bring-your-own LLM keys versus Intercom Fin's ~$0.99. For web-only products, Tiledesk or Chatwoot are the stronger open-source matches.

Are these open-source support platforms actually free to self-host?

Mostly, with caveats. Tiledesk, AnythingLLM, TGO, and ResolveKit are free to self-host in full. Chatwoot's core is free, but its Captain AI features are gated to Enterprise Edition even for self-hosted deployments. Dify is source-available under a modified Apache license with restrictions on operating it as a multi-tenant service - read the license before deploying commercially.

Can I bring my own LLM to these platforms?

Yes, for all seven. Most support OpenAI-compatible endpoints, which covers OpenAI, Anthropic, Gemini, and self-hosted models via Ollama or vLLM. ResolveKit routes every request through your own LLM API keys by default - nothing is proxied through ResolveKit's infrastructure unless you choose the managed tier.

What license do these projects use?

They range from fully permissive to source-available-with-restrictions: ChatterMate and ResolveKit's SDKs use Apache 2.0 / MIT, ResolveKit's backend and Tiledesk are AGPL-3.0 / open-core, Chatwoot's core is MIT with Enterprise AI features, AnythingLLM and TGO are open source, and Dify uses a modified Apache 2.0 license with commercial-use restrictions worth reviewing.

Get Started

Start Free - self-hosted ResolveKit is free forever under MIT/AGPL, no credit card required.