Pectus — self-hosted creator insights

./pectus.ai — last edit 2026-05-17written by Jesper Aström

Pectus is open-source scaffolding for data-driven apps. Build them. Compose them. Share them with the people you collaborate with. We're early. The contract works. The community is at zero. That part is up to you.

Ok. You're here and you're wondering what Pectus.ai is…

In short: a framework that makes building data-driven apps a lot easier by design. Inbound apps bring data in (Search Console, GA4, Meta, a CSV someone dropped in a folder). Outbound apps consume that data and produce something the user sees or publishes (a CMS, a dashboard, a stream overlay, a poster for LinkedIn). Skills are portable verbs that compose whichever apps are activated for a project. Add a new app, drop it in. Add a new skill, drop it in. Restart and it's live.

Think of it as the WordPress of AI-shaped apps, except the plugin layer is built and stitched by AI agents reading a small contract. The bundled apps (a CMS, GA4, Search Console, a few skills) are reference implementations. The point is the framework, not the apps that ship in the box.

Example: CMS

Let's say I want to build a CMS that tells me what content to produce every day, depending on shifts in demand, trends around me and what my competitors are writing on their websites? (This was the background to the content-insights which is an app that comes with pectus.ai base installation)

Inbound App: Connector that gets data from Google Search Console, Google Ads, Meta, GA4 (with potential to get from a ton more non free sites). Within it an interpreter that selects the most important data for what you want out of it, like rising keywords or simply ALL THE DATA.

Skill: I write a skill that can use a thorough process to analyse the data coming in and validate it and reformat it so that it can be used (do calculations etc).

Outbound App: A CMS that recommends 5 posts per day, connected with the brand and ICP as well as the updated market data coming in through the inbound app. The CMS can write blog posts, build landing pages, create ads and content for social media, with the right lighting, camera use and people, because of the APIs it's connected to and the camera descriptions that have been enabled in the brand context. And since it is all getting feedback data, it can, over time, explain to itself what is performing and what is not.

Example: Streaming service

Let's say I think that all available streaming services are too expensive and that OBS is too complicated to setup.

Outbound App: I simply build an outbound app that connects to YouTube, LinkedIn and all other services and streams to their platform. It will be built with artifacts such as bottom thirds, thumbnails, prerolls etc that are all on brand and in style.

Let's say I want to bring people's live comments into the stream.

Inbound App: I simply build an inbound app fetching the comments through the API.

Skill: I build a skill that delivers the comments to the Outbound App AND that filters any profanities.

Example: Performance Analysis

Let's say I want to create a performance analysis that looks at exactly why my performance has changed on my website, Facebook or Google Ads-account over the last period of time.

Inbound App: I build an inbound app that fetch ALL the data from ALL the platforms.

Skill: I build a skill that shows me everything that is actually contributing to or retracting from my growth (yes, it will require knowledge to write that skill, but it is very very very very possible. Bc. I just did it).

Outbound App: I build an outbound app that visualise this beautifully, on brand for each of my clients, but also depending on who is reading it. Some want more detail and some want only big pictures. The outbound performance app can do this.

Pectus provides the structure. It knows the brand and what the customers want. That's at pectus of everything. If your customer is a B2B-client or if it is a B2C-customer looking for trendy shoes. You're on top of your game!

In long: well, I had two fat problems.

Actually, it didn't start there. It all started in wanting to build an ad generator for one of my clients. Whilst exploring this I started thinking about two phat problems..

One was that branding always sucks in a digital context. The other was that everyone says they are data driven but no one really applies data to their work.

I wanted to solve both those problems.

I wanted to see what happened if instead of making "brand" an afterthought in a content management system or front-end code…. What would happen if we would build the content management system and front-end code from the brand.

I also wanted to see what happened if we put insights into the create loop. In plain text — what if instead of writing a report about what to create next, the system could write an abstract that a creator or AI-engine could use.

This was at the same time as OpenClaw started to make its rounds around the world.

Which got me thinking about the "self-optimizing website".

See. I don't like AI.

Especially not general model AI where you give an AI the job to do your job.

Not because I think the product is worse. Nah.

But because I see all competitive advantage from being different disappears as soon as you employ the same AI as anyone else. Which I've covered here when analyzing the "amazingness" of AI guided ad systems.

And then I got 100 other thoughts. And then and then and then… FUUUUUUU…..

So I was like: "what unites all of these thoughts?"

We need data to be interpreted or guided to be usable
One piece of data can be used in many ways
We therefore need a framework that can fetch and translate and then use the data in whatever way we want
And if there is something new introduced to the system it has to be able to adapt

And so… this is where the design of Inbound Apps and Outbound Apps and Skills (yeah, stole that one from Claude) was born… Inbound apps collect and interpret whilst outbound apps build and create. Skills help users create apps and build stuff that requires data.

I'm not a developer and this problem has most likely been solved before without me knowing. But the process collaborating with Claude Code, thinking of all of the different problems I usually face, I thought that I'd build a tool for me, release it for everyone to use…

So yeah. This became pectus.ai.

A framework where brand and ideal customer profile (ICP) are like kingpins guiding whatever comes out of it WHILST being sensitive to feedback necessary to make a company grow. Ie. whilst creating products, content, reports, tools, services built on data - it also receives feedback data that tells the apps if their product is being picked up or not.

Bc although a brand and a customer pain point might stick around for decades, what gives you attention and builds attraction changes. Ie. demand changes. So. In order to harvest demand - and in order to take a first step towards a self-optimising system… well… we needed a feedback model.

And since YOU are able to add to the interpretation of the data coming in from the feedback model YOU will always be able to outperform your competition IF you are smart enough. And that's why my framework is better than an AI autonomous one.

One more thing. Pectus is most useful when other people's apps run inside your install. The reference suite (a CMS, GSC, GA4, a few skills) shows what's possible. The value compounds when the suite gets longer because people other than me added to it.

Honest scope of what works today: web apps that render inside the Pectus admin, static sites published through the bundled Astro pipeline, inbound connectors that write to project-scoped Postgres tables, skills that call any LLM or API and persist structured output, multi-brand installs. The contract is small enough that an AI agent of your choice (Claude Code, Codex, Cursor, anything) can read it and write a conforming app. Most of the reference apps were built with Claude Code; other agents should work but nobody's stress-tested it yet.

What doesn't work today: mobile apps, multi-tenant SaaS, apps that need a different database, apps with their own auth model. The contract is also still moving — v0.4 to v0.5 may require small migrations in community apps. Watch the changelog.

Please get in touch at ja@pectus.dev if you bump into any errors. This is a moving train and there will be a few bugs as this baby grows up. Just ping me and I'll solve it. :)

The install lives at /install, the docs are at /docs, and the code is on github. — Jesper