Skip to main content
All issues

Monthly Dispatch: November 2026

Issue No. 77 min read

Your agent can now build software on Plexara and release it: an application with its own database, or a website on your own domain, with a person deciding what goes live. This month we moved plexara.io onto it.

Plexara is built in layers, and each layer is pressure-tested before the next one is built on it. Applications are made of automations, and automations hold the agent to the practices of a disciplined software team: drafts it can run without changing anything, tests that must pass before a save, releases, and rollback. Each capability reaches agents first, and the portal screens follow. Work runs from a person to their agent to Plexara: you say what you need, the agent builds it with Plexara's tools, and people review, approve, and use the result in the portal.

Because every capability is a tool an agent calls, a new frontier model receives all of it the day it connects. Anthropic's newest models are already at work on Plexara deployments, over the same open protocols as the models before them. Claude, ChatGPT, Gemini, and the ones that replace them get governed access to your data, the knowledge your team has recorded, and every tool below. The chatbot built into each vendor's product works on that one system, with that vendor's model.

What is new this month

Applications: systems your agent builds and releases

New

Describe the system you need and the agent builds it: a transactional PostgreSQL database of its own, automations that do the work, forms and endpoints for people and partner systems, and pages that open in a browser at the application's own address, for people who sign in through your identity provider or for the public behind a bot check. Each change to the database's structure is checked and tried on a test database first.

An application ships as a release that pins every part at a known version, so what runs is exactly what an administrator released. A saved change goes live at the next release, and a rollback puts the previous release's parts back in force. The Applications series walks through a shop an agent built with Plexara tools alone: a product catalog, a checkout that records the order and takes the stock in one step, an order confirmation, and a storefront. The same guide ships with every deployment, with the checkout as a ready-to-run example.

The Diagram tab of the shop application: boxes for the daily sales report, the sales dashboard, a webhook source and the order intake API, joined by arrows labelled reads, refreshes, event and writes to the tables in the application's database, with the Releases section below showing Export, Roll back and Release.
The diagram is drawn from the parts themselves: what each one reads, writes, calls and refreshes. Releases sit underneath, with Roll back beside Release.

Sites: websites on your own domain, built in dev and promoted

New

Every site has a private working copy called dev, and your agent changes dev only. You review dev at its own address, then promote that exact release through the environments other people use, such as staging and then production. When an agent asks to promote, roll back, or open a domain to the public, the decision goes to a person: in their AI client, or as a request in the portal's Approvals.

Every address that Plexara gives a site sits behind your organization's sign-in, and each environment is shared with the people or roles you name, so one site can serve a client preview, a department, or the whole company as an intranet. Only a domain added to the site can be public. Larger sites keep their content as records (products, posts, locations, people) and Plexara generates the pages from them. An existing site moves in from its source, and a change to a live site becomes a new dev release for a person to promote.

A site called marketing in the portal's Sites section. An explainer gives three steps: Changes land in dev, Promote one step at a time, and Only a domain you make public opens for anyone. Below it are three environment cards, dev, stage and prod, with prod marked Public domain and Requires approval, each with its current release and its addresses.
One site, three environments in promote order. dev is the private working copy, stage is where reviewers read it, and prod requires approval. Each card shows the release it serves and the addresses that reach it.

plexara.io now runs on Plexara

On November 13, plexara.io moved onto Plexara Sites on our own deployment. Its pages, lessons, Insights, newsletter archive, and more than 2,000 images are served by the same platform our clients use. The contact form is a Plexara form, and the facility map draws its tiles from our deployment. The site's layouts and navigation are templates the platform renders, so an edit made on the platform holds.

We run it under the same rules:

  • Agents work in dev. This month's lessons and Insights, and this issue's archive page, were written into dev by an agent and reviewed at a private address behind our sign-in.

  • People promote. Production requires approval, and since this week's release the platform holds every agent at dev. These pages reached plexara.io when one of us promoted them.

  • The rules live on the deployment. What our copy may say, and the checks to run before a promote, are a knowledge page on our deployment, and the site points every agent to it.

Deasil's other sites moved this month too. A marketing site, intranet, or partner portal that sits where your data and your agent cannot reach moves the same way: from its source, one environment at a time, with a person deciding what goes live.

Also new this month

Rules on the calls your agents make

New

Administrators can attach an automation to a connection so it runs on the calls other people make through it:

  • Policies decide before a call runs: allow it, refuse it with a reason, or hold it for approval. Allow ten refunds an hour, keep writes to business hours, or send any change to a supplier record to its owner.
  • Transforms change what comes back, in results and in exports. Card numbers or salaries can be masked for the people who should not see them, and those people cannot filter or join on the column to recover the values.
  • Watches start a run after a call succeeds, without slowing it: notify a channel when someone changes a price through Plexara, or record every write made through Plexara to a sensitive system.

When a call is held, the person's AI client offers to file a request, and an approver can grant an allowance, such as twenty more calls until Friday. Administrators can test a call against a connection's rules without reaching the service, and each call's audit record names the rules that ran. See Governance.

Smaller additions

  • Several triggers per automation. Automations run on demand, on a schedule, once per event (a webhook delivery, a form submission, or an approved email), and once per request (an endpoint). One automation can now hold several of each, and a new admin list shows every trigger in the deployment.
  • Diagrams and charts in the portal. D2, Graphviz, Excalidraw, Vega-Lite, and Lottie files display as drawings, charts, and animations, shared and commented on like any report.
  • Forms can be previewed by their owner and published to your signed-in people. Serving one to the public takes an administrator's approval of that version.
  • One library search across every report, file, and collection you can open.
  • API connections show which automations depend on them and flag when the service on the other end changes.

Thirty-four releases have gone out since the October issue, about one a day, and Deasil put each one on your deployment. Full notes are on the changelog.

From the Learning section

Four new lesson series since October, 28 lessons in all, and twelve new Insights.

Usage tip: ask for the first working piece of a long plan

Give your agent the whole goal and ask for a piece it can build this week:

Here is what we want this part of the business to look like in five years. Look at the data, the connections, and the automations we already have on Plexara. Tell me what we could build this week that moves us toward it, then build the first piece as an application: the database, the automations, and the pages. Show me the diagram and what the first release would contain before you release it.

The agent starts from what your deployment already holds: your data, your definitions, your connections. You review the diagram and the proposed release before anything goes live.

If there is a tool, a site, or a renewal you are rethinking, reach out and tell us about it. We read every message.

The Plexara team

Subscribe to the Plexara Newsletter

Once a month: new product features, MCP and AI educational resources, practical tips, and enterprise AI insights. Written for data leaders, engineers, and developers.

Powered by Buttondown. No spam. Unsubscribe anytime.