One app, many surfaces
One deployable app can provide chat turns, REST, MCP, multiple widgets, a main view, public content, events, jobs, and cron while consuming separately configured tools and services.
App surfaces ↗Capability comparison · August 2026 · sourced and dated
AI agent frameworks, runtimes, builders, and infrastructure compared.
Choose a category, then compare concrete capabilities among products that overlap there. KDCube appears in each category it implements; unlike products are not forced through one scorecard.
Verified August 2, 2026. Open any cell for its first-party source and scope note.
Capability matrices
Compare within a category
✓ Yes built in
Partial narrower, add-on, or separate
No explicitly outside the product
Not documented no first-party evidence found
KDCube distinctions
Capabilities that cross the category lines
Use these when your project needs more than one tab. Each item is a shipped combination, not a separate product you must assemble.
Native and hosted agents together
Use the native coding agent, host Claude Code, or preserve an existing LangGraph, CrewAI, Claude Agent SDK, or custom agent behind the same application boundary.
Agent runtimes ↗Per-agent capability and consent boundaries
Tools, skills, MCP, named-service operations, models, instructions, and subagents are bounded per agent; Connection Hub adds revocable per-user and per-agent grants.
Capability controls ↗Open-ended code behind a physical authority boundary
Generated code runs without network or platform credentials in the reference profile and can request governed tools through a trusted supervisor that keeps identity and secrets.
Execution isolation ↗Conversation, files, and user-controlled memory
Ordered conversation history, hosted files and artifacts, sparse workspaces, and evidence-backed memory that users can review, edit, and delete remain distinct but connected.
Memory reconciliation ↗Live app and agent control
Operators can switch an app version, settings, enabled surfaces, and access policy, while the Agent Forge stages models, instructions, tools, skills, event sources, and subagents before one apply.
Agentic configuration ↗Runtime services for app code
Apps receive deployment-neutral artifact storage and KV cache, transports, background work, request protection, outbound-event filtering, and bounded recording sinks.
App storage and cache ↗Product economics, not only cost charts
Attributable usage feeds plan quotas and prices, project budgets, wallets, internal or Stripe subscriptions, pre-run reservation, and actual-cost settlement.
Economics administration ↗Verification
How to read the ratings
Each category has its own peer set and feature rows. A rating is based on current first-party documentation or source code; each cell links to that evidence. “Not documented” is deliberately different from “No.” Partial covers add-ons, separate products, edition limits, or narrower contracts.
The matrix is re-audited monthly and after capability-changing KDCube releases. Corrections are recorded in the comparison changelog; the complete sourced KDCube contract remains in the capability ledger.
Open source · MIT licensed · Self-hosted
Bring your agent.
Keep LangGraph, CrewAI, Claude Agent SDK, or custom code — or start from the built-in agent. Add the production layer that fits your product.