Hostman Monthly Log:
August–September 2026

Hi from the Hostman team, and welcome to the first Hostman Monthly Log.

Around the turn of each month, we'll round up what's new across our services and what we've been up to. There's a lot to catch up on this time, so this first issue covers both August and September.

• Automatic setup and app cloning
• Agentic search, skills, and code execution
• A new look for the website and dashboard
• Markdown for your AI tools
• 8 new PostgreSQL extensions
• More updates
• Also shipped

Automatic Setup and App Cloning

Copy link

We've rebuilt the way you create an app. Now you connect your repository, and App Platform identifies the framework and runtime version, then fills in everything from the build and start commands to the health check path.

You get the final say: check what it found and change anything that looks off before the app is created. If you don't want to wait for detection, skip it and set things up yourself.

Also new: cloning. If you already have an app set up the way you like, clone it instead of starting from scratch. The copy keeps the same variables, commands, and settings, so there's nothing to re-enter. Your original app keeps running untouched, and the clone is a separate app, billed on its own. See exactly what gets copied.

Try it with your repo →

Agentic Search, Skills, and Code Execution

Copy link

AI Agents gained three new features. They're easiest to explain with one request: "Compare these three products from our catalog and calculate the cost of 12 units of each."

First, the agent has to find all three products. Before, it ran a single search and answered with whatever came back. With agentic search, it can search again, rephrasing the query as it goes, to get details on each product. New connections use agentic search by default, and you can switch existing ones in the agent settings.

Next, it needs to know how you compare products. Write that down once as a skill. The agent loads it only when a request calls for it. Every edit to a skill is saved as a new version, so you can always roll back to an earlier one.

Finally, the math. Language models aren't reliable at arithmetic, so the agent can now run Python or JavaScript in a sandbox, using a script you've attached to the skill or one it writes itself. Running code is free. 

Open your agents →

A New Look for the Website and Dashboard

Copy link

Hostman has a new look, built around a simple idea: the person using the cloud should come first. The redesign covers both the website, where you choose your infrastructure, and the dashboard, where you run it.

In the dashboard, we made the interface simpler and easier to navigate, so you can focus on your infrastructure. For example, we rebuilt the sidebar as one panel instead of separate cards. The account menu has moved into it from the header, and search itself has been reworked. Service pages are next.

When something needs a second pair of eyes, our support engineers are online 24/7, with a first reply in minutes on live chat.

Explore the new Hostman →

Markdown for Your AI Tools

Copy link

If you build with Cursor, Claude Code, or another AI coding tool, it probably reads our docs as often as you do. Until August, it had to work through HTML pages made for browsers. Now every docs page also comes in Markdown, and there's an index that tells your tool where to find things.

Add .md to any docs URL to get the page in Markdown, for example hostman.com/docs/app-platform/cloning.md. Or point your tool at llms.txt, which lists every section of the docs. 

Each section has its own llms.txt, so an agent setting up your Kubernetes cluster can load just the Kubernetes docs instead of everything. If you do want everything, it's all in llms-full.txt.

We also raised the rate limits on the docs, so an agent reading dozens of pages in a row is less likely to get blocked.

Browse the docs →

8 New PostgreSQL Extensions

Copy link

Your PostgreSQL clusters can now use eight more extensions, bringing the total to 21. Most of them take care of things you'd otherwise handle in code:

• Find what's nearby. earthdistance calculates distances between coordinates, so "stores near me" becomes a single query. It relies on cube, and the dashboard turns both on together.
• Search that forgives typos and accents. fuzzystrmatch matches words that are spelled slightly differently, and unaccent lets "cafe" find "café".
• Store categories, tags, and attributes. ltree keeps category trees and lets you query a whole branch at once. hstore stores key-value pairs in one column, and intarray speeds up queries on integer arrays, like lists of tag IDs.
• Skip the slow start after a restart. pg_prewarm loads tables and indexes back into memory, so the first queries don't crawl.

Here's typo-tolerant search in practice:

SELECT name
FROM products
WHERE levenshtein(unaccent(lower(name)), 'cafe latte') <= 2;

This finds both "Café Latte" and "Cafe Late".

To turn them on, open your cluster, go to the Configuration tab, and click Modify next to Extensions. 

Open your databases →

More Updates

Copy link

Run AI tools on your own server

With AI Agents, you build an agent from a model and your own instructions. The new AI Solutions section is different: it gives you ready-made tools you can run on your own server. Take OpenClaw. Pick it, and Hostman sets up a server with the agent already installed. You can chat with it right in the dashboard, no SSH needed. You'll also find n8n and Hermes there, they come preinstalled on a new server.

Create Kubernetes clusters faster and upgrade worker groups

We've rebuilt how Kubernetes clusters are created: the control plane is now ready in 3–5 minutes, and workers take about 2 minutes more. You can also move an existing worker group to a higher plan or a custom configuration instead of creating a new group. Its nodes restart one at a time while the change is applied. 

Set up redirects, headers, and compression for your app

Your app now has web server settings you can change in the dashboard, without touching your code. Redirect old URLs to new ones, so a domain move doesn't cost you search traffic. Add headers for CORS and security, like CSP and HSTS, and hide the ones that reveal your stack. Turn on gzip or zstd compression to make pages load faster, and for single-page apps, enable SPA fallback so links to any page open your app instead of a 404.

Mask sensitive fields in API responses

When you create an API token, you can now have it hide passwords, secrets, and private keys in API responses. They come back as hidden-by-api-key-policy instead, which makes the token safer to hand to scripts, CI pipelines, or AI agents. Existing tokens work exactly as before. See how to manage API keys.

Keep IP addresses with their project

IP addresses now live in projects, just like your other services. Move a service to another project, and its IPs go with it. Team members only see the IP addresses in projects they have access to, so no one can accidentally delete an address from a project they don't work on. Your existing IPs are already in the projects they belong to. See how projects work.

Also Shipped

Copy link

App Platform

• An app's environment variables can now be exported to a .env file in one click, and the limit is up from 100 to 1000 per app.
• Branch names can include non-Latin characters, as long as Git accepts them.

Managed Kubernetes

• New add-on: Kyverno checks, adjusts, or generates cluster resources based on your policies, so configurations stay within your security rules.
• New add-on: the cert-manager CSI Driver mounts certificates straight into pods, with no separate Certificate resources or Secrets.
• The add-ons list can now be filtered by category.
• Kubernetes 1.36.4, 1.35.8, and 1.34.11 are now available.

Managed Databases

• OpenSearch 3.7.0 is available for new clusters.
• Databases and database users now have statuses, so you can see whether a change is still in progress or already done.

Bare Metal

• 36 new server configurations, all billed by the hour with no minimum term.

AI Services

• Nine new text models joined the catalog, including GPT-6 Astra, Claude Fable 5.1, Gemini 3.8 Flash, and DeepSeek V4.1 Flash. Browse them all in AI Gateway's Models tab.

• Image generation got five new models: FLUX.2 [pro], [klein] 9B, and [max], plus GPT Image 2.5 Flare and Sunburst.

• Jev from TypeSafe is the first classifier in the catalog. Instead of writing a reply, it returns a category for the input text, so it's fast and cheap.

• The agent playground now opens right after you create an agent, with the chat in view instead of below the fold.

Dashboard and Servers

• SSH keys can be renamed and given comments. Each key's new Services tab lists the servers that use it, with their IP addresses, so you don't have to check servers one by one.

• Dashboard lists can now show 10, 20, 50, or 100 items per page.

Marketplace

• Six new networking images are available in the Cloud Servers Marketplace: 3X-UI, AmneziaWG, Headscale, Pritunl, Outline, and SoftEther.

That's it for this issue. Take a look around the updated dashboard, and if you need a hand, our support team is here 24/7.

See you next month for the next bunch of releases.

Open your dashboard →