ResolveKit vs Decagon vs Intercom: Which In-App AI Support SDK Actually Ships?
When engineering teams evaluate in-app AI support tools, the demos look similar. Chat interfaces, AI responses, some tool-calling capability. The differences that matter — SDK footprint, deployment model, operator control, and long-term maintenance burden — don't show up in screenshots.
This is a direct comparison of three platforms engineering teams actually consider: ResolveKit, Decagon, and Intercom's AI features. We'll look at what it takes to ship, what you give up with each, and which one won't become a maintenance headache six months from now.
How We Evaluated
We're looking at the same criteria engineering teams use when making a build-vs-buy decision for a production mobile app:
- Integration complexity: How long to first working prototype?
- SDK footprint: Impact on app size and startup time
- Native vs overlay: Does it feel like part of your app?
- LLM flexibility: Which models can you use?
- Operator control: What can the AI do without a human approving it?
- Deployment model: Cloud-only, self-hosted, or both?
- Maintenance trajectory: How much does this compound on your engineering team over time?
---
ResolveKit
Integration time: 2–4 hours to first prototype
SDK size: ~500KB iOS, ~1MB Android
Deployment: Self-hosted (MIT) or managed ($0.50/resolution)
ResolveKit is built for engineering teams that want full control over their support automation. The SDK is native (Swift/Kotlin), ships without a web view, and integrates directly into your app's UI layer. You bring your own LLM API keys — OpenAI, Anthropic, Ollama, or anything compatible with the tool-calling interface you've defined.
What makes it different: Operator approval flows are a first-class concept. Any tool function your agent can call — issuing a refund, canceling a subscription, resetting a password — can be gated behind an approval queue. Operators see exactly what the agent wants to do, who asked for it, and why, before anything executes.
The backend is open source (AGPL for server components, MIT for SDKs). If you want to self-host, you can. If you want managed infrastructure, you pay per resolution. No lock-in either way.
The catch: ResolveKit doesn't have a pre-built chat UI you drop in. You own the interface — ResolveKit gives you the session state, tool execution, and approval queue. If you want a polished chat UI out of the box, you'll need to build or choose a third-party component.
---
Decagon
Integration time: 1–3 hours for basic setup
SDK type: Native (iOS/Android) with hosted backend
Deployment: Cloud-only
Decagon positions itself as a low-integration option for teams that want AI support without operating infrastructure. Their SDK wraps a hosted agent backend, and the setup process is streamlined — connect your app, define your knowledge base, and the agent handles incoming questions.
The upside: Decagon is fast to integrate if you don't need deep customization. The hosted backend means less ops work upfront.
The tradeoff: You're constrained to Decagon's agent runtime and their supported tool integrations. If you need to call custom APIs, perform actions specific to your product, or implement nuanced approval logic, you're working within what Decagon exposes. LLM flexibility is limited to their supported providers. The operator dashboard is provided as-is — you configure, you don't extend.
For high-volume consumer apps with standard support flows (FAQ, order status, basic account changes), Decagon covers the job. For teams with custom workflows or data residency requirements, the hosted model becomes a constraint.
---
Intercom
Integration time: 1–2 hours (web SDK) / not native mobile
SDK type: Web widget (also available via Messenger SDK)
Deployment: Cloud-only, SaaS
Intercom is the established player in customer messaging. Their AI features (Fin, Resolver) are built on top of their existing Messenger platform — a web-based widget that lives on top of your app.
The upside: Intercom has brand recognition, a mature platform, and integrations with most major CRMs and helpdesk tools. If you're already in the Intercom ecosystem, adding AI support is incremental.
The fundamental problem for mobile teams: Intercom is a web overlay. On mobile, that means:
- Users leave your app to chat in a web context
- The agent doesn't have native access to your app's state
- Tool execution (refunds, account changes, etc.) requires your backend to expose APIs that Intercom can call — you build the bridge
- The experience feels like contacting support, not resolving an issue in-product
For web applications, Intercom is a reasonable option. For iOS and Android apps where you want support embedded directly into the experience, a web widget is a meaningful trade-down in UX and capability.
---
Side-by-Side Comparison
| | ResolveKit | Decagon | Intercom |
|---|---|---|---|
| iOS SDK | Native Swift (~500KB) | Native | Web widget only |
| Android SDK | Native Kotlin (~1MB) | Native | Web widget only |
| Web support | No (iOS/Android only) | Yes | Yes |
| LLM flexibility | Bring your own keys | Provider-limited | Provider-limited |
| Self-hosted option | Yes (MIT/AGPL) | No | No |
| Operator approval flows | First-class | Limited | Via third-party |
| Tool function support | Custom Swift/Kotlin | Predefined actions | Webhook/API |
| Approach to control | Operator owns the agent | Platform-managed | Platform-managed |
| Pricing | Free (self-hosted) / $0.50/resolution (managed) | Usage-based | Seat-based + AI |
---
What Actually Differentiates These Platforms
ResolveKit wins on control and extensibility. If your support workflows are custom — or will become custom as your product grows — ResolveKit gives you the building blocks without locking you into a managed runtime. The operator approval model means nothing happens without human oversight where you want it. The trade-off is that you own the UI layer.
Decagon wins on speed. If you need to ship a working prototype in a day and your support flows are standard, Decagon gets you there fast. The hosted model means less ops, less infrastructure, and less flexibility simultaneously.
Intercom wins on ecosystem. If you're already using Intercom for web support and want to extend AI capabilities across channels, the AI add-on makes sense. If you're building a mobile-native experience from scratch, Intercom's web-widget approach doesn't translate well.
---
The Question to Ask Before You Choose
The decision isn't really "which platform is best." It's "which platform will still make sense in 18 months when your support flows have changed, your product has new features, and your team has grown."
Support automation isn't a solved problem you configure once. It's a system that evolves with your product. The platforms that age well are the ones that give you control over the agent's capabilities, visibility into what it's doing, and the ability to course-correct without migration pain.
ResolveKit is built for teams that expect to grow into their infrastructure. Decagon is built for teams that want a solution that works today. Intercom is built for teams already invested in their messaging platform.
---
Ready to Build?
If you want full control over your in-app AI support agent — native SDKs, your own LLM, operator approval flows, and a deployment model that fits your infrastructure — ResolveKit is built for exactly that.