Run AI agents and entire experiences for customers without
praying your cost limits hold.
Bring LangGraph, CrewAI, Claude Agent SDK — or your own agent code. Keep your product logic. KDCube adds the production runtime around it: ordered conversations and streaming, user-scoped state, spend controls, isolated execution, and delivery from Git.
/plugin add kdcube-app-builder
$ npx kdcube-cli init
This page runs on KDCube.
The feed is one KDCube app: deployed from Git into a runtime bound to one tenant and project, and accounting for its model and tool use. Its output and execution telemetry show the application framework and production runtime working together. See the live apps →
Bring what you have. Add only what you need.
Build a complete AI app
Combine backend operations, multiple frontend surfaces, and one or many agents. Start with ready chat, streaming, web search, knowledge, memories, and files.
Build an AI app →Connect what you already run
Connect LangGraph, CrewAI, Claude Agent SDK, custom code, your UI, or backend services through webhook, REST, or conversation streaming.
Follow the LangGraph port →Configure data and services
Declare storage, properties, secret references, tools, APIs, identity, and user-connected services per app.
Connect a service with OAuth/OIDC →Deploy, update, and manage costs
Run locally, deploy apps from Git, scale workers, update independently, and track model, search, tool, and API use.
Set up and operate KDCube →Production concerns, handled together.
Real boundaries
One tenant/project · user-scoped workspaces · isolated code execution.
How isolation works →Connects both ways
Connect external accounts. Let approved agents and automation in.
How governed access works →Every call accounted
Spend attributed per user · app · turn. Budget checked before work starts.
How spend controls work →| Production concern | Your agent stack | + LangSmithplatform services | + KDCubeMIT · self-hosted |
|---|---|---|---|
| Keep existing agent code | ✓ already yours | ✓ supported | ✓ supported |
| Tracing & evaluations | add your choice | ✓ core product | runtime records · bring evals |
| Production agent hosting | build it | ✓ managed · Enterprise self-host | ✓ self-hosted runtime |
| Full application surfaces | build it | agent APIs · UI streams | ✓ UI · API · agents · jobs |
| Cost handling | build it | ✓ track & analyze | ✓ track & enforce |
| Generated-code execution | build it | ✓ sandbox capability | ✓ isolated executor · trusted tools |
| Accounts & delegated operators | build it | integrate in app | ✓ managed grants & credentials |
| License & deployment model | your choices | cloud · hybrid · Enterprise self-host | ✓ MIT · your infrastructure |
How KDCube runs your apps
Every surface enters through the same governed runtime. Ordered work reaches healthy workers, which load independent apps and run their APIs, agents, tools, and jobs.
Start in your terminal.
Clone · run locally · connect your agent · add one capability. Minutes, on your machine.
