2026
2026-08-02
KDCube application, agent, governance, and economics subsystem audit
Evidence correction
The KDCube capability ledger was expanded from 37 to 46 independently sourced contracts. The additions make complete subsystems visible rather than compressing them into generic labels: descriptor and UI agent configuration; request-time app settings and surface policy plus explicit code-version switching; the app transport map; multiple independently built widgets and a main view; discoverable public content; app storage and cache; per-operation CSRF; outbound event firewall and recording sinks; and plan, price, budget, wallet, subscription, payment, reservation, and settlement administration.
The niche matrices now project those contracts only where they answer that category's buyer question. The application-builder matrix compares managed agent configuration, UI and public surfaces, live client streaming, MCP consumption and publication, hot app controls, app and user configuration/secrets, storage, access policy, event governance, and economics. The self-hosted and managed-runtime matrices compare the narrower runtime and operating contracts. No all-vendor scorecard or dedicated vendor-versus-KDCube section was introduced.
The source audit checked every matrix URL. One moved n8n log-streaming page was replaced with its current first-party path; the remaining links resolved. Desktop tables now divide the available width among only the peers in the selected category, while mobile retains a sticky capability/KDCube pair and horizontal peer scrolling.
Why it matters: readers can project a concrete use case onto comparable products without losing the larger KDCube product contract or mistaking terse table labels for the implementation boundary.
46-capability KDCube ledger → · Category-specific matrices → · Economics model → · Agent configuration → · Live app configuration →
2026-08-01
Niche-specific benchmark and separate KDCube capability ledger
Evidence correction
The old universal score grid was withdrawn. It put application runtimes, agent frameworks, visual builders, managed services, model endpoints, observability products, voice services, SDKs, and deployment infrastructure under one scoring model. That produced category errors, stale labels, and unsupported negative verdicts.
The current page has one canonical, switchable comparison. Its six niches are Agent frameworks, Self-hosted agent runtimes, Managed agent runtimes, Application builders, Observability and evaluation, and Deployment platforms. Every niche owns an independent peer set and independent criteria, so a framework is not scored as an incomplete runtime and infrastructure is not described as an agent platform.
The complete source-backed KDCube inventory now lives on a separate 46-capability ledger. It preserves the app SDK and lifecycle, provider and consumer surfaces, native and hosted agents, shared harness, chat/UI/data/event/job contracts, tools/skills/MCP/named services, Connection Hub delegation, generated-code isolation, retrieval and user-reconcilable memory, cache governance, accounting, economics, monitoring, analytics, scaling, and deployment boundaries. Each card keeps the shipped contract, primary source, audit date, and limiting boundary together.
The self-hosted-runtime niche contains KDCube, Trinity, OpenHands, Agno AgentOS, and Letta Server as peers for that one decision; no vendor receives a dedicated comparison section. LangSmith, Langfuse, and Arize Phoenix appear only in the observability/evaluation niche. A concise KDCube section names the integrated contracts that are materially unusual, then links to the full ledger rather than repeating it.
Several KDCube underclaims are now explicit. User-controlled durable-memory reconciliation is supported. The native agent writes and executes code, while the reference split profile places generated code in a separate networkless, no-credential container and brokers governed tools through a trusted sibling. Native-agent subagents and hosted framework agents are both represented. App-scoped provider/consumer surfaces are distinct from each agent's tools, skills, MCP, and named-service inventory. Connection Hub owns platform identity, OAuth/OIDC connected accounts, delegated credentials, per-agent grants, and consent.
MCP is no longer compressed into a generic "two-way" label. The framework and application-builder niches separately compare Consume MCP and Publish MCP, including how each path is configured and whether publication is native, supplied by an attached server product, or requires a separate MCP server. For KDCube, consumption is configured in the app connection registry and narrowed per agent; publication is an async @mcp provider surface with descriptor-owned visibility and authentication.
Deployment is no longer summarized as a singular container or as Kubernetes-only: the page distinguishes CLI-managed Docker Compose, Kubernetes/Helm, and the descriptor-driven AWS ECS/Terraform topology whose deployment implementation is currently private. The decision guide explains what to choose and what can be combined; the benchmark method states the niche rule, evidence policy, date scope, and correction cadence.
The agent-framework hosting guide and Bedrock AgentCore runtime deep dive were also aligned with current primary sources. Historical release notes remain below as dated research history; they are not presented as the current snapshot.
Why it matters: the benchmark answers the buyer's category question first, while the separate capability ledger preserves the full KDCube audit. Product breadth is not collapsed, and unlike technologies are not forced into one favorable-looking score.
KDCube capability ledger → · Connection Hub → · KDCube MCP → · LangGraph → · Trinity → · LangSmith → · AgentCore →
2026-07-13
Comparison table re-audited as of today
Position holds
Full table re-audited as of today; confirmed current through KDCube release 2026.5.22.442. Added two comparison rows across every tab — virtual / long-running task jobs (off-turn) and neutral tenant-scoped artifact storage — reflecting the operational-layer primitives that shipped in the May 22 release. No competitor feature changes were observed since the May 22 snapshot.
Why it matters for the comparison: the new rows make the shared-operational-layer axis explicit in the table rather than only in the deep-dives. Bedrock AgentCore is scored Partial on both (per-agent-runtime execution billed separately; S3/bucket storage that isn't tenant-scoped by the runtime); every other alternative leaves off-turn job lifecycle and neutral artifact storage to whatever you stitch on the side. Nothing in the competitor field moved to warrant re-rating an existing cell.
KDCube release notes →
2026-05-22
KDCube release 2026.5.22.442 — virtual task jobs + neutral artifact storage
Position holds
Two new platform primitives that deepen the operational-layer pitch: virtual task job execution (long-running off-turn bundle work where the runtime owns the lifecycle and the bundle owns the work) and a neutral bundle artifact storage API (so artifacts that aren't tied to a turn — long-form documents, generated PDFs, scheduled-job outputs — get the same tenant scoping as everything else without each bundle inventing its own backend). The CLI also split kdcube init from kdcube refresh, with --tenant/--project leading as primary form.
Why it matters for the comparison: reinforces the "shared operational layer" axis the compare deep-dives already lean on. Bedrock AgentCore charges per agent runtime and storage; LangGraph leaves long-running work to whatever you stitch on the side (Temporal/Hatchet); CrewAI's orchestration is in-turn. KDCube's bundle gets virtual jobs + neutral artifact storage shared with the rest of the fleet at flat marginal cost.
Release notes →
2026-05-21
KDCube release 2026.5.21.145 — native-agent protocol surface tightened
Position sharpens
The May 21 release repositions timeline_text as a first-class mid-turn channel (it was a fallback path), gives each action inside a multi-action thinking block its own streaming key, and tightens the exploration→exploitation causality rule with explicit runtime guards. Stream families and longrun lifecycle are now first-class in the SDK docs. The chat.complete path no longer overwrites the streamed answer with the raw final_answer.
Why it matters for the comparison: this is the native-agent protocol differentiation moving from "implicit in the runtime" to "explicit in the SDK contract." Per-action streaming keys and first-class mid-turn channels cannot be expressed through a provider-native tool-call shape alone; they use the timeline as shared state instead of the provider's message log. The fleet-of-apps and operational-layer axis was already the lead in compare.html; this release makes the agentic-protocol axis more visible alongside it.
Release notes →
2026-05-13
KDCube release 2026.5.13.117 — off-turn job substrate landed
Position holds
The May 13 release finalized the substrate for off-turn bundle work: @cron now carries a span dimension (system | process | instance) gated by canonical enabled.cron.<alias> flags; a proc-owned scheduler with per-job Redis locks and a reconcile loop landed alongside selectable scheduler backends; and a new bundle background job stream (kdcube_ai_app/infra/jobs/stream.py) gives off-turn jobs a generic async progress/log channel addressable from the admin UI and the chat surface. Bundle props now reactively call an on_props_changed hook; dynamic per-resource config overrides (expr_config, tz_config) ship with an in-place admin editor; feature gating is uniform across bundle/api/mcp/widget/cron via canonical enabled.* flags.
Why it matters: the same substrate now supports long-running app workloads such as memory analysis and reconciliation. The current memory posture is explicit and user-reviewed: durable memories are inspectable records, snapshots protect restore paths, and reconciliation runs as a tracked async job. The detailed comparison is in Claude Dreams-style memory vs. KDCube user memory.
Release notes →
2026-04-21
Memory maintenance direction
Model clarified
KDCube's memory model is being clarified around separate surfaces: native-agent conversation memory, internal memory anchors, indexed notes, durable user memory, turn announcements, and snapshot-backed reconciliation. Durable user memory is not an automatic transcript rewrite; it is user-visible state with scope, provenance, and restore paths.
Why it matters for KDCube: this avoids mixing agent-authored internal notes with durable user preferences. The next meaningful product work is not a hidden "dream" pass; it is a widget-centered workflow for search, pinning, snapshots, async analysis, proposal review, apply, and restore.
Comparison post →
2026-05-10
DIY column — refreshed for the 2026 stack
No impact
The "Build yourself" column was rewritten to reflect what teams actually stitch in 2026: Temporal and Hatchet have largely replaced Celery Beat for durable agent scheduling; Pydantic AI v1.85+ sits alongside LangGraph as a serious lightweight agent loop with first-class durability and Logfire-native tracing; Langfuse (acquired by ClickHouse in January 2026) is now a widely used open-source observability option; managed sandbox options expanded to include Northflank and Daytona alongside E2B and Modal (Firecracker / microVM-backed isolation is non-trivial to self-host).
Why it matters: KDCube's value is the integration of these layers. The DIY path's individual primitives keep getting better; what stays painful is the cross-layer plumbing — tenant-aware authz, hot-reload of bundles, atomic budget enforcement, and channeled streaming over Socket.IO + SSE + REST.
Temporal → · Hatchet → · Pydantic AI → · Langfuse →
2026-05-10
AWS Bedrock AgentCore — five-meter consumption pricing
Gap closes
AgentCore (GA late 2025; April–May 2026 wave added São Paulo, Frankfurt, Tokyo regions) now bills across five separate consumption meters: Runtime (vCPU + GB-hr/sec), Gateway (per-MCP-op + per-indexed-tool), Memory (per event + per record retrieved), Identity (per token issued), and Policy ($0.000025/auth request) — plus S3 for browser artifacts. Production-grade Browser runtime, Code Interpreter, and Memory (short + long-term) reached GA in the same wave.
Why it matters for KDCube: AgentCore continues to mature on the runtime axis, but the five-meter cost surface is hard to forecast and tightly coupled to AWS-only operational primitives (Gateway, Memory, Identity, Policy, CloudWatch). The "self-host on infrastructure you own with one budget surface" pitch holds; the "managed services lock you in" framing on Why is the right read.
AgentCore pricing → · AgentCore product page →
2026-05-10
CrewAI v1.14 + Enterprise on-prem — directly competitive
Gap closes
CrewAI v1.14.x (April 2026) shipped checkpoint TUI with fork/lineage, reasoning-token + cache-token accounting, and A2A (agent-to-agent) protocol docs. CrewAI AMP — the proprietary control plane on top of the MIT engine — now offers an on-prem option, with real-time dashboards, per-agent cost tracking, RBAC, SOC2, SSO, secrets, and PII detection.
Why it matters for KDCube: CrewAI is the closest competitor on the "open core + on-prem control plane" axis. Differentiation: KDCube's runtime is fully MIT (no Enterprise tier gate), bundles are surface-agnostic (chat / REST / iframe / MCP / @cron in one app), and pricing is your infrastructure cost — not "contact us." Watch this row.
CrewAI changelog → · AMP pricing →
2026-05-10
Anthropic Agent Skills — open packaging standard
No impact
Anthropic's Agent Skills standard (open-sourced December 2025) was adopted by 32 tools within months — VS Code, ChatGPT/Codex, Gemini CLI, Junie, Kiro, Goose — and Vercel's skills.sh marketplace now lists ~89k skills. It is a packaging standard, not a runtime: a Skill is a folder of prompts + scripts + metadata, not a hosting environment.
Why it matters for KDCube: KDCube bundles can wrap Skills — the two layers compose rather than compete. We may add a future row noting compatibility, but Skills doesn't change the Why-page positioning between libraries / managed services / self-hosted runtime.
The New Stack — Agent Skills →
2026-05-01
KDCube — comparison table re-audited
Position holds
Snapshot refreshed against the May-2026 KDCube state: split-executor strategy, hardened isolated execution runtime, open-source frontend runtime config served from ingress, Tier 1 app docs, and file-backed app authority. KDCube's scoring on Governance & Security and Economics columns was unchanged; the generated-code isolation row gained a stronger primary source.
Why it matters: the next external snapshot of KDCube vs alternatives won't surprise readers — the differentiating rows are still the same ones (per-tenant budgets, atomic admission, tenant-scoped audit, isolated execution).
KDCube changelog → · Engineering write-up: split executor & trusted supervisor →
2026-02-11
OpenAI — formal AGENTS.md definition
No impact
OpenAI's developer post formally defined AGENTS.md as a cross-tool standard, building on the AGENTS.md spec released August 2025 and now stewarded by the Linux Foundation's Agentic AI Foundation. Adopted by OpenAI Codex, Cursor, Factory, Claude Code, and Gemini CLI.
Why it matters for KDCube: not a comparison-table row — but it's the file every coding agent reads first. KDCube added AGENTS.md at the website repo root in May 2026 to be discoverable to the same tool ecosystem.
developers.openai.com — Custom instructions with AGENTS.md →
