Notable changes to the Hivedeploy platform. Subscribe to the RSS feed for updates.

2026-07-18 — Public reliability ledger

Added

• proof.hivedeploy.in — our published reliability ledger: raw and converged pass rates from real deployment campaigns, every failure root-caused with a public evidence dossier, costs included. Numbers are labeled with their exact envelope (agent × cloud × scenario set × gate version) — no unqualified claims.
• Docs: per-cloud deployment troubleshooting (GCP, AWS) written from those campaign dossiers, and a new self-healing explainer.

2026-07-16 — AWS application deployments

Added

• Full Java application topology envelope on AWS: Fargate HA, EKS, ASG + ALB, 3-VM EC2, Lambda SnapStart — mirroring the GCP envelope
• Zip-source deploys build images via CodeBuild inside your account
• Observability discovery for ECS, ASG, and EKS deployments; EKS metrics fall back to node-level EC2 series when Container Insights isn’t installed, so panels are never structurally blank

Fixed

• Health model no longer counts ALB-generated 502/503/504 pages as “app up” — verification waits for real application responses
• Availability-zone capacity errors (InsufficientInstanceCapacity) now rotate to a sibling AZ automatically

2026-07-14 — Health verification v1.1 + Cloud Run probe fix

Changed

• Health verification now gates on health + logs + metrics; backing-store connection is measured and displayed, non-gating (the failing class proved environmentally dominated, not app-defect dominated)

Fixed

• Cloud Run: Google’s front end reserves /healthz externally (404 to outside callers even when the container serves it) — the public probe now falls back across candidate health paths, unblocking the entire Cloud Run class
• Container images are pinned by digest at deploy time, eliminating stale-tag races

2026-07-02 → 07-12 — Self-healing deployments

Added

• Bounded auto-repair ladder: deterministic fixers first (capacity rotation, name-collision rename, JVM memory hardening, template hygiene, and more), then AI repair with anti-repeat steering, then redeploy — at most 5 attempts, every one visible in the run log
• Learned fixers: the platform distills new deterministic fixers from its own repair history and applies them without an AI call
• Kubernetes deployments get correct startup/liveness/readiness probes; zone fallback on GCP capacity errors; JVM apps on small instances are auto-hardened against OOM crash-loops

2026-06-13 → 06-28 — Generative deployments, the flywheel, and the learning curve

Added

• Generative deployments: ask for a topology outside the template catalog and the agent plans, renders, and deploys it — behind the same gates and health verification as everything else
• Template promotion flywheel: every verified generative success is captured, quarantined, and ratified into the deterministic template library — the next identical ask routes through the frozen template (faster, cheaper, variance-free)
• Learning curve: generative mistakes are distilled into lessons that steer future generation (guidance first, hard gate once proven)
• Quota preflight: deployments check cloud quota headroom before applying — you fail in seconds with the exact limit named, not after 20 minutes of churn
• Cross-agent isolation: fixes for one agent/cloud are scoped and tested to never regress the others

2026-05-15 → 05-17 — Multi-cloud parity + RBAC

Added

• Multi-cloud deployment pipeline parity across AWS / GCP / Azure adapters
• Provider-native IAM permission sets per deployment topology
• RBAC admin UI: deny-by-default roles, per-agent gating, approval hierarchy

2026-05-13 — GCP WIF integration

Added

• GCP cloud accounts now connect via Workload Identity Federation instead of service-account impersonation
• Required APIs documented in Connect GCP (sts.googleapis.com, iamcredentials.googleapis.com, cloudresourcemanager.googleapis.com)
• Org ID banner with copy button on the GCP connect form
• OIDC issuer endpoints at /.well-known/openid-configuration and /.well-known/jwks.json
• API-not-enabled detection in the probe — error message now tells you the exact gcloud services enable command to run

Fixed

• Probe error messages now include the underlying GCP response body for faster diagnosis
• owner_id in probe path now correctly uses the org_id (was a hardcoded literal that prevented the WIF binding from matching)

Breaking

• Legacy SA-impersonation flow removed (no customers were on it at flip-over time)

2026-05-12 — Public signup + access strategy

Added

• SIGNUP_MODE env var with three modes: open, invite_only, waitlist
• /signup page adapts UI based on the active mode
• /verify-email page for email confirmation
• /admin/waitlist page for admins to approve waitlisted users
• Cloudflare Turnstile captcha on signup
• Velocity-breach detection that auto-logs CRITICAL when signup floods exceed threshold

Security

• Pre-merge correctness gate covering 49 unit + 14 adversarial + 300 property-fuzz tests across the signup/auth/admin domains
• All velocity breaches surfaced to operator inbox for manual response (auto-flip-to-invite-only deferred to a later release)

2026-05-08 — Notifications page redesign

Changed

• New /notifications page UI: mono micro-labels for metadata, sans-serif for titles, time buckets (Today / Yesterday / Earlier this week / Earlier)
• Pill-chip filter bar with unread count badge
• Empty / error / loading state precedence
• Wrapped in shared AppTopbar + new MarkAllReadButton in the actions slot
• Conditional render of “Clear filters” CTA only when filters are actually applied (no no-op buttons)

Fixed

• Focus-visible ring restored on filter chips (a11y regression from the shadcn migration)
• Retry path uses a nonce to avoid leaking cancel tokens across retries
• Mono font now scoped to labels/timestamps only — titles render in the platform’s body sans, matching the rest of the app

Internal

• Bell dropdown UI is unchanged in this release; only the dedicated /notifications page was redesigned

2026-04-30 — Plan picker on post-verification

Added

• Post-verification redirect to /onboarding/choose-plan
• Plans visible: Free, Starter, Pro (per Billing)
• Stripe Checkout integration for paid plans
• Free-plan detection derived from price, not SKU id (more robust to catalog renames)
• Onboarding wire payloads now strictly typed (dropped http<any>)
• Subscribe button is full-width on mobile

Internal

• Mock-Stripe harness for local-dev signup flow testing

2026-04-29 — Billing limits + overage markup

Added

• Per-plan credit budgets with soft limits + hard caps
• Real-time credit consumption tracking per org
• BILLING_OVERAGE_MARKUP_DEFAULT env var (default 1.40 multiplier)
• Stripe metered-usage reporting wired to deployment + agent-time consumption

Operations

• PAYMENT_PROVIDER=none bypass for self-hosted operators not using Stripe

2026-04-28 — Source persistence + Connection + Logs tabs

Added

• Connection details persisted across deployment sessions (no more re-entering the same params)
• Connection tab on deployment detail page
• Logs tab streaming live Terraform output (and historical logs after apply completes)
• Source code import for coding agents (Python, Node.js, frontend) — scope-limited to coding deployment agents, not infrastructure agents

2026-04-25 — Frontend agent: per-framework deployment

Added

• Per-framework specialist memory for the frontend agent — Next.js, React SPA, Vue, Angular, Astro, SvelteKit each have their own deployment template
• Cloudflare Pages, Vercel, AWS Amplify, GCP/Azure Static Sites, S3+CloudFront targets all wired

2026-04-15 — Specialist agents wave 1

Added specialist agents for:

• Language backends: Python, Java, Node.js, Go, Rust, .NET, PHP, Ruby
• Databases: Postgres, MySQL, MongoDB, ClickHouse
• Caching & Search: Redis, Elasticsearch
• Vector & AI: Qdrant, vLLM (beta — frontend mock only, backend specialist arriving in wave 2)
• GitOps: ArgoCD (beta)

Each agent owns the full lifecycle (deploy + monitor + scale + tune) rather than being a one-shot deploy bot. See Use the agents for per-agent how-to’s.

2026-04-08 — Multi-cloud foundation

Added

• Cloud account abstraction supports AWS (cross-account IAM role + external ID), GCP (service-account impersonation, later replaced by WIF), and Azure (App registration + service principal)
• Provider-agnostic data model — adding a new cloud is a single adapter implementation
• Per-account connection probe + verification flow

2026-04-01 — Initial preview release

The Hivedeploy orchestrator’s first preview build. Domain-expert agents, gated deployment flow, multi-cloud foundation.