Added Claude Haiku 5.5 and GPT-6.1 Sol to the AI App Builder model picker and AI Providers. Haiku 5.5 is now the default Anthropic model for platform-managed AI.
AI Builder
Highlights
Anthropic Claude Haiku 5.5 (claude-haiku-5-5) is the newest Haiku, built for fast, high-volume work. It costs $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens.
Haiku 5.5 replaces Sonnet 5 as the default Anthropic model for platform-managed AI, and is available on the Free plan.
OpenAI GPT-6.1 Sol (gpt-6.1-sol) is the newest Sol, at the same price as GPT-6 Sol ($2 input and $10 output per million tokens).
Both are selectable for builder sessions and bring-your-own-key AI providers, with cost tracking updated.
Why it matters
Managed AI usage costs far less per request, so plan token allowances go further. Sonnet and Opus remain one click away in the model picker for harder tasks.
Builds now follow your package manager and Node version
Generated Dockerfiles install with pnpm or yarn when your repo uses them and pick the Node version from your repo, instead of always using npm on Node 20.
Deployments
Highlights
pnpm and yarn. If the repo has pnpm-lock.yaml or yarn.lock, the build installs with that tool, so peer dependencies resolve the way they do on your machine.
Node version. NEXUS AI reads .nvmrc, .node-version, engines.node, the @types/node version, or your CI Node version, and builds on Node 20 to 24 accordingly.
Fixes. Projects such as Vite apps that failed on a missing peer dependency or needed Node 22 now build.
Configurable container memory on Pro
Pro and Enterprise plans can raise a deployment's RAM from the default 512 MB to up to 4 GB, so memory-heavy apps that build or run at startup no longer get killed.
Deployments
Highlights
Set memory per deployment. Pass memoryMb (512 to 4096) when you deploy from source on the NEXUS AI Docker provider, through the API or the nexusai_deploy_source MCP tool.
Pro and above. Free and Starter plans keep the 512 MB default. Asking for more returns a clear upgrade message instead of failing later.
Stays with the deployment. Redeploys and restarts reuse the memory you set.
Why it matters. Apps that run a build when the container starts, or load large models, are killed with exit code 137 when they run out of RAM. A larger limit fixes that.
Community templates: deploy tested open-source apps in one click
A new Templates page lists open-source apps we have deployed and tested, starting with AnythingLLM, OmniRoute, NEXUS AI Gateway, and Flixty, each with a one-click Deploy button and setup steps.
Deployments
Highlights
Templates page. Browse and search tested open-source apps at nexusai.run/templates, filtered by category.
One click per app. Each template carries the exact deploy link we tested, including the Dockerfile, port, and settings it needs.
Setup that is honest. Every page lists what gets created, what you need first, the steps after deploy, where data lives, and any security step, such as setting a password.
Launch templates. AnythingLLM, OmniRoute, NEXUS AI Gateway, Flixty, TREK, and open-connector, with more added as we test them.
Secrets without links. Deploy links can ask for a secret on the deploy page (require=), generate one at deploy time (generate=), and fill in the app's own URL ({url}), so apps that need an admin token or encryption key deploy locked down from the first start.
Deploy button: Dockerfiles in subfolders, custom ports, and link settings
The Deploy to NEXUS AI button can now build a Dockerfile outside the repository root and route to any port, so existing open-source projects deploy without changes.
Deployments
Highlights
Dockerfile path. Point the button at a Dockerfile such as docker/Dockerfile with dockerfile in nexus.json or dockerfile= on the link. The repository root stays the build context.
Port. For Dockerfiles without EXPOSE, set port in nexus.json or port= on the link. NEXUS AI declares it for you, and redeploys keep it.
Settings on the link.env=KEY=value sets variables for repositories without a nexus.json. The deploy page lists them for review and editing, and names that change how a process starts are refused.
Deploy repositories you do not own. Link options need no files in the repository, for example ?repo=https://github.com/Mintplex-Labs/anything-llm&dockerfile=docker/Dockerfile&port=3001&env=STORAGE_DIR=/app/server/storage.
Safer installer rule. Dockerfiles may now pipe curl or wget to a shell when the download is https, which common installers such as uv and bun require. Plain http stays blocked.
Use your own editor with the AI Builder: live two-way sync in the CLI
nexus builder dev keeps a local folder in live sync with an AI Builder project, so VS Code, Cursor, or any editor works as your IDE while you keep chatting with the AI.
AI App Builder
Highlights
Live two-way sync.nexus builder dev --project <id> --dir ./my-app --open downloads the app, opens it in VS Code or Cursor, and keeps it in sync: saved edits become checkpoints, and AI changes appear in the folder.
Chat in the same terminal. Prompts at the dev> prompt run AI turns on your latest code; pending edits are saved first.
Merges instead of rejections. Edits to different parts of a file from you and the AI are combined. Lines changed on both sides get git-style conflict markers, which are never uploaded; the file syncs once you resolve them.
One-shot sync.nexus builder sync runs a single merge pass for scripts, and upgrades folders created with pull.
Safe by default. A folder that suddenly loses most of its files is not mirrored as mass deletions.
Why it matters
Before, pull and push were one-time copies: any AI change after a pull made push refuse, so you had to choose between your editor and the AI. Now you can use both on the same app at the same time.
Deploy to NEXUS AI button for GitHub READMEs
Add a one-click Deploy to NEXUS AI button to any public GitHub repository, with an optional nexus.json that describes the databases, worker, and settings the app needs.
Deployments
Highlights
README button. Paste one line of markdown and anyone can deploy your public repository to their own NEXUS AI account.
nexus.json. Declare Postgres, MySQL, MongoDB, or Redis, a background worker, a root folder, build and start commands, and environment variables: fixed values, secrets generated at deploy time, or values the person deploying must enter.
Deploy page. Visitors see the repository and exactly what will be created, sign up or sign in, and return to the same page to deploy. Secret values are entered there, never put in the URL.
Sign-in keeps your place. Signing up or signing in from any link now returns you to the page you were on.
Why it matters
One-click deploy buttons are how many developers try an app for the first time. Open-source projects can now offer that on NEXUS AI, with the database and worker wired up.
Deploy one folder from a repo with several apps
A new root directory setting lets you deploy the backend/ or frontend/ folder of a repo that holds more than one app, from the CLI, the dashboard, or MCP.
Deployments
Highlights
Root directory.nexus deploy source --root-dir backend, the Root Directory field in the dashboard deploy forms, and the rootDir field on the nexusai_deploy_source MCP tool build only that folder of the repo or upload.
Works on every target. NEXUS AI, Google Cloud Run, AWS App Runner, and Azure Container Apps all honor it, and in-place redeploys keep the folder you chose.
Clear errors. A folder that does not exist, is a file, or points outside the repo is rejected before the build starts.
Why it matters
Full-stack repos often keep the frontend and the API side by side. Automatic detection picks only one of them, and which one depended on the target. Now you name the folder and deploy each app on purpose.
Deploy into your own AWS account and Azure subscription
Apps and managed databases now deploy into the AWS account or Azure subscription you connect in Deployment Providers, and saved cloud credentials are encrypted at rest.
Deployments
Highlights
Your AWS account. When you add an AWS provider with your own access keys (Pro and above), new App Runner apps and RDS databases are created in that account and billed to it. NEXUS AI sets up the ECR repository, build bucket, CodeBuild project, and IAM roles it needs there.
Your Azure subscription. When you add an Azure provider with your own service principal (Pro and above), new Container Apps and Azure Database servers are created in that subscription and resource group.
Each deployment remembers its account. Stopping, scaling, deleting, custom domains, logs, database snapshots, and redeploys always act in the account a deployment or database was created in. Deployments made before this release keep running where they are.
Google Cloud prefers your project. When you have added your own Google Cloud provider, it is used ahead of the platform-managed one.
Encrypted credentials. AWS secret keys, Azure client secrets, and Google Cloud service-account keys saved in Deployment Providers are now encrypted with AES-256-GCM.
Why it matters
Teams that need their apps, data, and cloud bill in an account they control can now get that on AWS and Azure as well as Google Cloud, without changing how they deploy.
The AI Builder agent now works with every AI provider
OpenAI, Google Gemini, xAI Grok, OpenRouter, and custom OpenAI-compatible providers now build apps as an agent that reads files, makes targeted edits, and verifies its work, the same way Claude does.
AI Builder
Highlights
Agent mode on every provider. The builder's agent (list and read files, targeted edits, npm install and type checks in the App preview sandbox) previously ran only on Claude. It now runs on OpenAI, Google Gemini (with native function calling), xAI Grok, OpenRouter, and custom OpenAI-compatible endpoints, including the Free plan's default model.
Graceful fallback. If a model cannot form tool calls reliably, that message finishes as a single response instead of failing.
Predictable usage. On the Free plan with a platform-provided model, each message runs up to 8 agent steps and 150K tokens. Paid plans and your own API keys allow up to 12 steps and 300K tokens.
Why it matters
Whichever model you pick, the builder now edits only what needs to change and checks its work, so fewer changes break the preview and follow-up requests stay small.
Undo, redo, and discard in the AI Builder code editor
The AI Builder code editor now has Undo and Redo buttons, a Discard changes button for unsaved edits, and an Undo action after every save.
AI Builder
Highlights
Undo and Redo buttons in the editor toolbar step through your edits to the open file, the same as Cmd/Ctrl+Z and Cmd/Ctrl+Shift+Z. Each file keeps its own history, including across saves.
Discard changes throws away a file's unsaved edits in one click, and Undo brings them back.
Undo after saving. Saving, creating, or deleting a file shows a confirmation with an Undo button that restores the project to how it was just before that change.
Jumping to a syntax error puts the cursor on the error's line without scrolling the page sideways.
Why it matters
Trying a change in the code editor is safe: you can step back one edit at a time, drop everything you haven't saved, or undo a save you regret.
Upload files to the AI Builder with syntax checks, HTML preview, and AI fixes
Upload source files into a builder project or start a new project from them. Each file is type-detected and syntax-checked, HTML sites preview instantly, and the AI can fix reported syntax errors.
AI Builder
Highlights
Upload into a project or start a new one. Click Upload in the builder header, drop files anywhere on the builder, or choose "upload files" on the AI Builder start page.
Type detection and syntax checks. JavaScript, JSX, TypeScript, TSX, CSS, JSON, HTML, SVG, XML, and YAML are checked in the browser before anything is saved, with line and column numbers. HTML checks include inline scripts and styles and stray, misnested, or unclosed tags.
Instant HTML preview. A plain HTML site previews in the builder with its local CSS, JavaScript, and SVG files, including links between pages. HTML files can also be previewed in the upload dialog before they are added.
Fix with AI. "Add and fix with AI" asks the AI to fix only the reported syntax errors in the uploaded files, then checks them again.
Limits. Up to 20 files per upload, 500 KB per file, and 2 MB per upload. Text source files only.
Why it matters
Existing code no longer has to go through GitHub or chat to reach the builder. You see problems before they break the preview, and each upload is a version you can revert.
New models: Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, Grok 4.7
Added Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 to the AI App Builder model picker and AI Providers.
AI Builder
Highlights
Anthropic Claude Opus 5.5 (claude-opus-5-5) is the newest Opus, built for long-running agentic coding and knowledge work. It costs less than Opus 5 ($4 input and $20 output per million tokens).
OpenAI GPT-6 Sol (gpt-6-sol) brings GPT-6 quality at a lower price than GPT-6 Astra.
OpenAI GPT-6 Luna (gpt-6-luna) is the fastest and cheapest GPT-6 model, suited to small edits and high-volume work.
xAI Grok 4.7 (grok-4.7) is xAI's most capable model for coding and agents.
All four are selectable for builder sessions and bring-your-own-key AI providers, with cost tracking updated for each.
Why it matters
Teams can move to the newest model from each provider as soon as it ships, and pick a cheaper tier (Opus 5.5, GPT-6 Sol, GPT-6 Luna) without giving up current-generation quality.
Documentation built for AI agents
A new site at agents.nexusai.run gives AI agents Markdown docs, a one-file llms-full.txt, an OpenAPI spec with real schemas, and an MCP tool catalog with the scope each tool needs.
MCP
AI agents now have documentation written for them at agents.nexusai.run.
What changed
Markdown docs for agents. 32 short documents cover deployments, managed databases, SQL safety rules, backups, storage, secrets, access tokens, plans and limits, errors and rate limits, and more. Browsers and search engines get the same pages as HTML.
One-file corpus.llms.txt indexes every page, and llms-full.txt holds all of them in a single fetch.
OpenAPI with real schemas. The GPT-compatible API has full request and response schemas, and the most-used platform routes list their request fields.
MCP discovery. The tool catalog lists all 74 MCP tools with the OAuth scope each one needs, and /.well-known/mcp.json describes the server before a client connects. An RFC 9727 API catalog is at /.well-known/api-catalog.
New models: Claude Fable 5.1, GPT-6 Astra, Gemini 3.8 Flash
Added the latest frontier models from Anthropic, OpenAI, and Google to the AI App Builder model picker and AI Providers, with Gemini 3.8 Flash as the new managed default for Google.
AI Builder
Highlights
Claude Fable 5.1 (claude-fable-5-1) is now selectable for builder sessions and bring-your-own-key AI providers. It is Anthropic's most capable model, for demanding reasoning and long-horizon agentic work.
OpenAI GPT-6 Astra (gpt-6-astra) is available as the newest GPT option.
Google Gemini 3.8 Flash (gemini-3.8-flash) is available and is now the managed default for Google, including on the Free plan. It also leads the Gemini flash overload-fallback chain.
Cost tracking, plan allowlists, and the builder model menu were updated for all three.
Why it matters
Teams can pick the newest model from their preferred provider without waiting on a separate rollout, and Free-plan builds get Google's latest flash model by default.
Solutions pages by role, and a redesigned public navigation
Added dedicated Solutions pages for seven audiences, from developers to vibe coders, and rebuilt the public site navigation with grouped dropdown menus.
Website
Highlights
Added /solutions/developers, /solutions/founder, /solutions/sales, /solutions/product-managers, /solutions/small-business, /solutions/entrepreneur, and /solutions/vibe-coders, each covering how NEXUS AI fits that audience from first prototype to deployed product.
Each page ships prerendered content, per-page metadata, FAQ and WebPage structured data, and responsive hero imagery for search engines and AI answer engines.
Linked the new pages from the Solutions hub, the site footer, and the sitemap.
Rebuilt the public header with grouped Product, Solutions, and Resources menus, keyboard and screen-reader support, and a reorganized footer.
Why it matters
Someone evaluating NEXUS AI for a specific role now lands on a page written for that role, and the reworked navigation makes the rest of the platform easier to find.
Support channels by plan and a Free plan usage policy
Added a public customer support page, moved Starter and Pro onto ticket and phone support, and published a Free plan usage policy that scopes the Free tier to testing and development.
Billing
Highlights
Added a public customer support page covering how to open a ticket from the dashboard and from Claude or ChatGPT over MCP, with copy-and-paste prompts and the four support tools.
Starter now includes email support and support tickets. Pro adds phone support on top of email and tickets. Free stays on community support.
Published a Free plan usage policy: the Free plan is for testing, development, and evaluation, and is not intended for commercial or production business applications.
Added both pages to the footer, sitemap, prerendered SEO metadata, and llms.txt.
Why it matters
The support path a plan includes is now stated on the pricing page and reachable from a page of its own. The Free plan keeps its full feature set for evaluation, with a clear line for when a workload should move to a paid plan.
Configurable retention for scheduled database backups
Scheduled database backups now default to a 7-day retention window and prune automatically, with the option to raise it for longer history.
Databases
Highlights
Scheduled backups for a database service now default to a 7-day retention window, set automatically the first time auto-backups are enabled.
Retention can be raised for longer history, but never lowered below 7 days, so every service keeps a safe minimum recovery window by default.
Backups past their retention window are pruned automatically as part of the existing hourly backup job. Backups you upload yourself are never auto-deleted.
Set retention with nexus db backup-schedule <service-id> --enable --retention <days>, the nexusai_db_backup_schedule MCP tool, or the backup schedule API.
Why it matters
Auto-backups no longer accumulate indefinitely and eat into disk quota. Every service keeps at least a week of recovery history out of the box, and teams that need more can opt into it explicitly.
Import an existing GitHub repository into the AI Builder
Bring an existing repo into the AI Builder instead of starting from a blank prompt: any framework gets a real, running preview, then you can keep editing or deploy it.
AI Builder
Highlights
Added "Import an existing GitHub repository" to the AI Builder's start page: pick a repo your GitHub App installation can access, choose a branch, and NEXUS AI clones it, detects the framework, and seeds a new builder session with the real files.
Preview is no longer Next.js-only: Next.js and Vite apps get an instant hot-reload dev container, and every other supported framework (Django, Rails, Laravel, FastAPI, and more) gets a real running preview in a container instead of just a static file view.
Deploying an imported session now uses its actual detected framework instead of assuming Next.js.
Fixed page navigation inside the preview: clicking a link in an imported app no longer escapes the preview and lands on the main NEXUS AI dashboard.
Fixed the repository picker defaulting to the same repo on every import — it now requires an explicit selection, and added a "Manage GitHub App access" link so a different repo can be granted access without leaving the builder.
Why it matters
Teams with an existing codebase can bring it straight into the Builder's live preview and chat-driven editing instead of starting over from a prompt, regardless of what framework it's written in.
New comparison and use-case landing pages for search and AI answer engines
Added dedicated landing pages for platform comparisons and common deployment use cases, each with FAQ and HowTo structured data, and fixed a routing issue that was serving AI crawlers an error on blog pages.
Website
Highlights
Added dedicated pages for Railway, Vercel, Render, and Replit alternatives, plus use-case pages for deploying AI-generated apps, one-prompt deployment, one-click deployment, full-stack deployments, MCP-driven deployment, and enterprise vibe coding.
Each page includes a structured comparison table where relevant, FAQ and HowTo schema, and a breadcrumb trail.
Fixed an nginx routing bug that was returning a 401 to AI crawlers requesting blog pages.
Why it matters
People, and AI assistants answering on their behalf, searching for a specific comparison or use case now land on a page built for that exact query instead of the general homepage.
New comparison and use-case landing pages for search and AI answer engines
Added dedicated landing pages for platform comparisons and common deployment use cases, each with FAQ and HowTo structured data, and fixed a routing issue that was serving AI crawlers an error on blog pages.
Website
Highlights
Added dedicated pages for Railway, Vercel, Render, and Replit alternatives, plus use-case pages for deploying AI-generated apps, one-prompt deployment, one-click deployment, full-stack deployments, MCP-driven deployment, and enterprise vibe coding.
Each page includes a structured comparison table where relevant, FAQ and HowTo schema, and a breadcrumb trail.
Fixed an nginx routing bug that was returning a 401 to AI crawlers requesting blog pages.
Why it matters
People, and AI assistants answering on their behalf, searching for a specific comparison or use case now land on a page built for that exact query instead of the general homepage.
Deploy pipeline build reliability fixes
Fixed several build-time failure modes in the container build pipeline, including repo-shipped Dockerfiles that assumed a package-lock.json and volume mount option handling.
Deployments
Highlights
Repo-shipped Dockerfiles that assume a package-lock.json no longer fail the build when the repository does not have one.
Fixed volume mount option handling in generated container configuration.
Why it matters
Fewer builds fail on container configuration details that had nothing to do with the application code itself.
Managed database redeploy reliability fixes
Fixed an issue where redeploying a database-backed app in place could leave the database using an out-of-sync password, and repeated deploy-from-source calls could create a duplicate deployment instead of updating the existing one.
Databases
Highlights
Database passwords now resync automatically on every in-place redeploy instead of only when a stored value is missing, fixing connection failures that could follow a redeploy.
Password-sync outcomes are now visible in build logs for easier troubleshooting.
Repeated deploy_source calls (CLI, API, or MCP) for the same app now match and update the existing deployment by name instead of creating a duplicate.
Why it matters
Redeploying a database-backed application, or re-running a source deploy, no longer risks a broken connection string or a stray duplicate deployment.
Bucket file uploads via MCP, and multi-CIDR managed database access
Added an MCP tool for uploading files directly to a storage bucket, and expanded managed database network access to support multiple allowed CIDR ranges.
MCP
Highlights
New MCP tool nexusai_bucket_file_upload lets an MCP-connected agent upload a file directly to an S3-compatible bucket.
Managed database network access now accepts multiple CIDR ranges instead of a single one, for teams that need to allow more than one office, VPN, or peered network.
Why it matters
Agents can now move files into bucket storage without leaving the conversation, and managed database access can match real-world network topologies with more than one trusted range.
Dedicated Enterprise page and private infrastructure content
Added a dedicated /enterprise marketing page covering on-premises, private cloud, and hybrid deployment for NEXUS AI Enterprise, along with supporting blog content on private AI infrastructure and the AI App Builder.
Website
Highlights
Added a dedicated /enterprise page covering on-premises, private cloud, and hybrid deployment models, with links from the main navigation, footer, and sitemap.
Published supporting long-form content: a deep dive on private AI infrastructure and a deep dive on how the AI App Builder actually works under the hood.
Why it matters
Teams evaluating NEXUS AI for regulated or infrastructure-sensitive workloads have a single, dedicated page to see exactly which deployment models are available, and where data and infrastructure control actually sit.
Claude Opus 5 in the AI Builder and AI Providers
Added Claude Opus 5 to the AI App Builder model list and AI Providers, alongside the existing Claude model lineup.
AI Builder
Highlights
Claude Opus 5 is now selectable in the AI App Builder's model menu and in AI Providers.
Priced the same as Claude Opus 4.8, with no separate plan-gating changes required.
Why it matters
Teams that want Opus-tier reasoning for more complex generation and edits can select it directly, without waiting on a separate rollout.
Deployment pipeline security hardening
Hardened the deployment pipeline against injection vulnerabilities in repository cloning and generated compose configuration, and added a startup safeguard against default encryption secrets.
Security
Highlights
Repository cloning for deployments now runs without a shell and blocks internal, loopback, and link-local/metadata network addresses, closing a command-injection and SSRF path in repo and branch input.
Fixed a YAML-escaping gap in generated compose configuration that could let a crafted storage mount path inject an unintended configuration entry.
Production now refuses to start if core encryption secrets are left at their default, placeholder values.
Why it matters
These changes close injection paths in deployment inputs and turn a misconfigured secret into a hard startup failure instead of a silent security gap.
Vibe-code in Claude, preview in the AI Builder
New MCP tools let Claude and other MCP clients push generated apps into the AI Builder for a live preview before deploying, and pull the files back to keep iterating in chat.
AI Builder
Highlights
New MCP tool nexusai_builder_push: an MCP client (Claude, Cursor, ChatGPT, and others) pushes a generated file set into the project's AI Builder session and returns a preview link.
Opening the link shows the app running in the builder's live preview instantly, with no infrastructure spun up and no cost.
New MCP tool nexusai_builder_pull: the client reads the current files back, including edits made in the builder UI, so iteration continues in chat.
Every push is a restorable checkpoint in the session history, labeled with its origin.
Deploy from either side: the builder's Deploy button or the existing MCP deploy tools.
Why it matters
Until now, deploying was the only way to see an app generated in Claude chat actually running. The builder handoff gives vibe coders a free, instant preview step between chat and production.
AI App Builder: click-to-edit element selection
Click any element in the live preview and tell the AI what to change about it. No more describing DOM structure in words.
AI Builder
Highlights
A "Select element" toggle on the live preview highlights elements as you hover and captures the one you click.
The selection appears as a chip in the chat composer; your next instruction applies to that exact element.
Works together with image attachments: click the broken element and paste the mockup in the same message.
The selection stays fully local: the preview runs in a sandboxed iframe and only a short element description is shared with the AI.
Why it matters
"Make THIS button bigger" now works by pointing at the button instead of describing where it lives in the layout.
AI App Builder: image attachments the AI can see
Attach design mockups or bug screenshots to builder chat messages. The AI sees the images and uses them to build the app or fix visual issues.
AI Builder
Highlights
Attach up to 3 images per message (PNG, JPEG, WebP, GIF) via a picker or by pasting from the clipboard.
The AI sees attached images through vision and treats them as design references to match or bug screenshots to fix.
Images from earlier turns stay visible to the AI, so "match the screenshot I sent earlier" works.
Images are downscaled in the browser before upload, keeping requests small and responses fast.
The chat composer was redesigned in the style of modern AI chat interfaces: inline attach and send controls, Enter to send.
Why it matters
Describing a design in words is the hardest part of building visually with AI. Now you can just show it: paste the mockup and say "make it look like this."
AI App Builder: push to GitHub
Added the ability to push the current builder snapshot to a GitHub repository, reusing the organization's existing GitHub App installation.
AI Builder
Highlights
Added "Push to GitHub" to the builder, using the same GitHub App connection already used for GitHub-based deployments.
Supports pushing to a new repository or an existing one.
Each push creates a new commit containing the complete current snapshot, including deletions.
Why it matters
Apps built conversationally in NEXUS AI can graduate into a real, version-controlled GitHub repository without setting up git or copying files by hand.
AI App Builder: full-stack apps, built-in auth, and self-healing preview
Extended the AI App Builder to generate full-stack Next.js and Prisma apps with an auto-provisioned managed database, built-in email/password authentication, and automatic recovery from build errors.
AI Builder
Highlights
The builder can now generate full-stack Next.js + Prisma applications, not just React single-page apps.
Deploying a full-stack build automatically provisions a managed Postgres database and wires up the connection string; the live preview renders the same app against mock data.
Added a built-in email/password authentication contract (hashed passwords, signed session cookies) that the builder can add to an app on request.
The preview now self-heals: build errors trigger up to two automatic AI-driven fix attempts before asking you to step in.
If a deploy fails after launch, a "Fix with AI" action repairs the code and redeploys, reusing the same managed database instead of provisioning a new one.
Why it matters
Builder-generated apps can now include real user accounts and a real database, and common build or deploy failures are resolved automatically instead of leaving you to debug generated code by hand.
AI App Builder: chat, live preview, and deploy
Launched an in-dashboard AI app builder: describe an app conversationally, watch it render in a live local preview, edit the code directly, and deploy when it looks right.
AI Builder
Highlights
Added a chat-driven app builder at /projects/:id/builder, reachable from "Build with AI" on any project.
Every turn renders a live preview in a fully local, sandboxed iframe. No code or file contents leave the browser during preview, and there is no third-party sandbox or CDN in the loop.
Generated code is editable in place; manual edits create a new checkpoint just like an AI turn.
Every assistant turn stores a complete, restorable snapshot of the app, so you can revert to any earlier version from the version menu.
Deploying from the builder reuses the same deployment pipeline as the rest of the platform, including provider and region selection.
Why it matters
You can go from a plain-language description to a running, deployed application without leaving the dashboard, while keeping full visibility and control over the generated code at every step.
Marketing site SEO and AI crawler discoverability
Hardened the crawlable prerendered output of the marketing site with consistent structured data, an accurate sitemap, and explicit support for AI crawlers.
Website
Highlights
Expanded prerendered marketing pages with consistent titles, canonical tags, and JSON-LD structured data.
Added FAQPage schema and comparison tables to alternative/competitor pages.
Made the sitemap dynamic, with real last-modified dates for blog and knowledge-base content.
Confirmed AI crawlers (GPTBot, ClaudeBot, anthropic-ai, PerplexityBot) are explicitly allowed and served correct content.
Why it matters
Search engines and AI answer engines now index accurate, page-specific content across the marketing site instead of a single homepage summary, improving discoverability for people evaluating NEXUS AI.
Landing page enterprise refresh
Replaced the homepage hero with a live animated deploy terminal, added a trust strip, and refreshed typography and metadata for a more credible first impression.
Website
Highlights
Replaced the AI-art hero slideshow with an animated fake-deploy terminal showing a real nexus deploy flow end to end.
Added a trust strip highlighting the MCP Registry listing, tool count, supported clouds, and AES-256-GCM encryption.
Introduced a dedicated display typeface for headings and refreshed page titles and social preview metadata.
Why it matters
The homepage now demonstrates what NEXUS AI actually does in the first few seconds instead of relying on generic imagery, and shares metadata that matches what the product delivers.
Admin-managed changelog
Added a public changelog page, super-admin changelog CRUD, published-only public APIs, sitemap inclusion, and sidebar navigation.
Platform
Highlights
Added a changelog_entries data model and migration.
Added public published-only changelog API endpoints.
Added super-admin CRUD for changelog entries.
Added /changelog to the public site, footer, sitemap, and prerender output.
Added /admin/changelog to admin routing and sidebar navigation.
Why it matters
Product updates can now be managed from the admin UI and published directly to a public changelog.
Billing updates: Stripe customer repair and refund policy
Added automatic repair for stale Stripe customer IDs during checkout and published a refund policy page for billing transparency.
Billing
Highlights
Checkout now verifies saved Stripe customer IDs and recreates missing or deleted customers when needed.
Added a public refund policy page and linked it from the legal footer.
Added the refund policy route to public SEO and prerender metadata.
Why it matters
Billing setup is more resilient when Stripe mode/account data changes, and customers have a clear public policy for refunds and cancellations.
Full-stack app deployments from the UI
Added UI support for deploying full-stack applications, expanding dashboard workflows beyond single-service deployment actions.
Deployments
Highlights
Added full-stack application deployment workflows in the UI.
Expanded deployment forms and API client support for multi-part app submissions.
Improved project and deployment screens used during full-stack deployment flows.
Why it matters
Teams can launch more complete applications from the dashboard without dropping into manual deployment tooling.
Azure Database provider
Added Azure Database for PostgreSQL/MySQL Flexible Server support with private VNet, private DNS, Container Apps integration, and point-in-time restore.
Cloud
Highlights
Added Azure Flexible Server, network, and private DNS SDK support.
Added Azure Database provider support with private VNet and Azure Container Apps integration.
Added Azure Flexible Server snapshot/restore through point-in-time restore.
Added resource-provider registration and region-scoped VNet reuse behavior.
Why it matters
NEXUS AI now supports managed database workflows across AWS, Google Cloud, and Azure, matching the platform goal of multi-cloud deployment operations.
Private managed database networking and proxy access
Added private database networking support, including VPC-mode RDS connectivity and an in-VPC/VNet HTTPS-to-SQL proxy for query access.
Databases
Highlights
Added private VPC-mode RDS connectivity for AWS App Runner deployments.
Added an in-VPC/VNet SQL proxy for private managed database access.
Added query API support for public direct and private proxied managed databases.
Tightened MCP managed database query enforcement so read tools remain read-only and writes require manage scope.
Why it matters
Private managed databases can be operated through NEXUS AI without forcing public network exposure for routine query and management workflows.
GCP Cloud SQL managed databases
Added Google Cloud SQL as a managed database provider with Cloud Run connector support, backup handling, and clearer API-disabled errors.
Cloud
Highlights
Added the Cloud SQL Admin API dependency and GCP Cloud SQL provider adapter.
Added Cloud Run to Cloud SQL connectivity using native connector behavior.
Added backup-run handling and provider-specific error messages.
Pinned Cloud SQL settings to supported editions and transport behavior.
Why it matters
NEXUS AI expanded managed database support beyond AWS, giving Google Cloud users a first-class database path for deployed applications.
Managed database attachments and provisioning reconciliation
Added deployment-time environment injection for managed DB connections plus background reconciliation for provisioning and snapshot state.
Databases
Highlights
Injected managed database connection variables into deployments at deploy time.
Added scheduled reconciliation for database provisioning and snapshot states.
Added validation for managed database create and restore inputs.
Improved cleanup behavior to reduce cloud resource leaks.
Why it matters
Managed databases became operationally safer and more usable because apps can receive connection settings automatically, while background jobs keep long-running cloud operations in sync.
Managed databases: AWS RDS launch
Introduced the managed database foundation with schema, RBAC, plan limits, REST endpoints, MCP tools, CLI commands, and AWS RDS provisioning.
Databases
Highlights
Added the managed database schema and provider adapter interface.
Added AWS RDS provisioning, status, delete, snapshot, and restore support.
Added REST endpoints for create, list, delete, attach, snapshot, snapshots, and restore.
Added MCP tools and CLI commands for managed database workflows.
Added plan limits, RBAC permissions, and audit events for managed database operations.
Why it matters
Teams can provision and attach production database infrastructure from the same control plane they use to deploy applications.
Mobile and public site refresh
Updated public pages, mobile responsiveness, docs, landing content, and core marketing/resource surfaces for a cleaner user experience.
Website
Highlights
Improved mobile presentation across public and application pages.
Updated docs, homepage content, footer links, resources, FAQ, and public positioning.
Added richer content around deployment workflows, alternatives, and production use cases.
Why it matters
The product became easier to evaluate from mobile devices and clearer for teams comparing NEXUS AI with traditional deployment platforms.
NEXUS AI MCP server published to the official registry
Published the NEXUS AI MCP server to the official Model Context Protocol registry, making it discoverable and installable directly from MCP-compatible clients.
Verified OAuth discovery metadata resolves correctly for registry-driven client setup.
Documented the registry listing alongside existing manual setup instructions.
Why it matters
Users of Claude Desktop, Claude Code, Cursor, Codex, and other MCP-compatible clients can now find and connect to NEXUS AI directly from the client marketplace instead of copy-pasting configuration.
Security and SEO hardening
Improved security posture with dependency overrides and auth fixes while expanding public SEO, sitemap, and structured metadata coverage.
Security
Highlights
Added dependency overrides for known frontend security advisories.
Improved security-related fixes across auth and platform surfaces.
Added richer BlogPosting JSON-LD and sitemap lastmod metadata.
Updated public documentation and content surfaces for better search and AI crawler discovery.
Why it matters
The platform became safer to operate and easier for customers to discover through documentation, search, and AI-indexed public content.
MCP, OAuth, and fine-grained access scopes
Expanded the MCP platform with fine-grained OAuth scopes, token presets, registry metadata, and verification tooling for agent-driven deployments.
MCP
Highlights
Mapped MCP tools to fine-grained scopes for safer automation.
Added grouped scope selection and access token presets.
Added OAuth discovery metadata for expanded scopes.
Added MCP registry metadata, validation targets, and smoke-test scripts.
Updated documentation to reflect the real OAuth endpoints and scope model.
Why it matters
AI agents can perform deployment work with narrower permissions, making NEXUS AI safer to use from Claude, ChatGPT, Cursor, Codex, and other MCP-compatible tools.
Deployment stop/start data safety
Fixed stop and start so they preserve database containers, mounted volumes, and bucket configuration instead of rebuilding from scratch, and added automatic recovery after a host reboot.
Deployments
Highlights
Stop now performs a soft stop: containers, attached volumes, and network configuration are preserved.
Start rehydrates the existing deployment configuration instead of rebuilding the image from scratch.
Added boot reconciliation so deployments that were running before a host reboot come back automatically.
Delete remains a full teardown, including volumes and reserved ports, when that is explicitly requested.
Why it matters
Stopping a deployment to save quota, then starting it again, no longer risks losing database data, mounted storage, or bucket access. The same protection applies automatically if the underlying host restarts.
Container and database observability
Added runtime metrics, container metrics history, database metrics dashboards, and stronger operational visibility across deployed services.
Observability
Highlights
Added container metrics collection and history views.
Added database metrics and query-history surfaces.
Improved operational dashboards for monitoring deployed services.
Expanded documentation and SEO surfaces for monitoring and deployment workflows.
Why it matters
Teams can diagnose application and database behavior from the NEXUS AI control plane instead of stitching together separate monitoring tools for basic visibility.
Persistent storage volumes and object buckets
Added storage workflows for persistent volumes and S3-compatible buckets, including attach/detach operations and documentation for production usage.
Storage
Highlights
Added persistent volume workflows for deployment-mounted storage.
Added object bucket management with S3-compatible credentials.
Added attach, detach, usage refresh, credential rotation, and file operations for storage resources.
Added documentation for storage setup and backup/restore scenarios.
Why it matters
Applications can now keep durable files and bucket-backed assets across redeploys instead of relying only on ephemeral container filesystems.