Open source vs Intelligence
What ships in CopilotKit's open-source core versus what's part of the CopilotKit Intelligence layer — Threads, persistence, inspection, and self-hosting.
CopilotKit ships in two layers: an open-source frontend stack (npm packages, no external service required) and CopilotKit Intelligence, which adds saved conversations, realtime sync, cloud-hosted inspection, and an admin console. Intelligence extends the open-source packages. The SDK and the runtime stay the same. The boundary is which capabilities you unlock.
The 30-second version#
- The open-source packages are enough for building a working agentic application with framework-native APIs, prebuilt chat, frontend tools, gen-UI, and any AG-UI integration. The license is MIT. The packages run anywhere.
- CopilotKit Intelligence adds threads, persistence across reloads and devices, realtime WebSocket sync, inspection, and an admin console.
- CopilotKit Intelligence runs two ways — as Cloud-hosted CopilotKit Intelligence, or self-hosted via the
copilot-intelligenceHelm chart on your own Kubernetes. - Your app code stays the same for open source only, cloud-hosted, and self-hosted. The boundary is a runtime config switch.
| Open source only | Needs CopilotKit Intelligence | |
|---|---|---|
| Agent chat, frontend tools, agent state, gen-UI | Yes | |
| Any AG-UI integration (LangGraph, Mastra, Built-in, custom) | Yes | |
| Conversation history that survives reloads / devices | Yes | |
| Frontend thread API + realtime sync across tabs / devices | Yes | |
| Inspector, admin console, multi-tenancy | Yes | |
| Thread inspection and admin console | Yes |
What's in the open-source core#
Everything you need to build a working agentic app, with no external service from CopilotKit:
- Frontend SDK —
@copilotkit/react-core, exporting hooks and prebuilt components for chat, tools, state, threads, and generative UI.
- Runtime —
@copilotkit/runtime, the request handler that mounts in your application server. Framework-agnostic core plus thin adapters for Express, Hono, Bun, Deno, Cloudflare Workers. - AG-UI protocol — open standard for agent ↔ frontend communication. 16 standardized event types, transport-agnostic.
- Built-in Agent —
BuiltInAgentruns an in-process agent loop (OpenAI / Anthropic / Google / any AI-SDK model) without requiring an external framework. - Integrations — first-party adapters for LangGraph (Python / TypeScript / FastAPI), Mastra, CrewAI, Pydantic AI, Microsoft Agent Framework, AWS Strands, Google ADK, and others.
- Generative UI primitives — tool rendering, state rendering, headless APIs, and framework-native theming.
MIT licensed. Run it on any host, in any cloud, with any LLM provider.
What's in CopilotKit Intelligence#
A separate backend service that sits beside your runtime and unlocks the capabilities that need durable, multi-tenant infrastructure to be honest about:
- Threads & Persistence — durable, resumable conversations that survive page reloads, span devices, and sync in realtime via WebSocket. Backed by Postgres + Redis with single-writer locks on active runs.
- Inspection & history — project-scoped thread history, event timelines, admin views, and API key management in the Intelligence web app.
- Inspector & Admin Console — a web UI for browsing thread history, inspecting agent traces, rotating API keys, and managing projects and organizations.
- Multi-tenancy — three-level isolation (organizations, projects, users) so a single deployment can host production and staging side by side without one reading the other's data.
Cloud-Hosted vs Self-Hosted#
Same CopilotKit Intelligence — you pick where it runs:
- Cloud-hosted CopilotKit Intelligence is the service CopilotKit operates. Sign up, get an API key, and point your runtime at it. Pick this when you want zero ops on the platform side.
- Self-Hosted installs the
copilot-intelligenceHelm chart into your own Kubernetes cluster. Pick this for data residency, on-prem requirements, or compliance control. You operate the workload (Postgres, Redis, ingress, OIDC, secret rotation); the chart and the same CopilotKit Intelligence image power the install.
Application code doesn't change between cloud-hosted and self-hosted — same frontend SDK and runtime APIs. Only the platform endpoint and credential set change.
Where the boundary lives#
The open-source runtime knows how to talk to CopilotKit Intelligence. It does not require Intelligence.
If your app keeps state in your own backend, the open-source packages may be enough. When you want CopilotKit to store the conversation across users, devices, and sessions, use CopilotKit Intelligence.
Licenses#
The open-source SDKs and runtime use MIT. CopilotKit Intelligence and Showcase have separate terms.
Where to go next#
- Open-source quickstart — Quickstart gets a working agentic app running in ~10 minutes against the Built-in Agent. No platform credentials needed.
- CopilotKit Intelligence overview — CopilotKit Intelligence lists every platform capability and the per-feature setup.
- Cloud-hosted setup — Cloud-hosted CopilotKit Intelligence when CopilotKit operates the service.
- Self-hosted deployment — Self-host CopilotKit Intelligence for the Helm-chart install on your own Kubernetes cluster.
- Architecture deep-dive — CopilotKit Intelligence architecture covers runtime/platform roles, project boundaries, threads, and realtime sync.