Content Velocity
Twenty AI jobs draft pages, layouts, themes, summaries, keywords, and alt text — humans review instead of producing from scratch
Platforms & Infrastructure
The Content Platform Where AI Agents Do the Work — and You Stay in Control
PressGraph is a graph-native agentic CMS. Twenty built-in AI jobs draft content, design themes, compose layouts, and optimize your internal linking — every suggestion reviewed by you, every run fully audited. Runs on standard PHP hosting.
SEO
Native topical clustering, hub scorecards, link-health audits, hreflang, sitemaps, and structured data build measurable topical authority
GEO & AEO (AI Search Visibility)
A clean content graph with typed relations, canonical structure, and semantic markup gives generative and answer engines exactly the structured context they reward
Governance & Compliance
Suggestion queues, full AI audit trails with cost tracking, mandatory revisions, and consent-gated events make AI adoption defensible to clients and regulators
What it's about
PM PressGraph CMS
PressGraph rethinks the CMS from first principles. Instead of posts, pages, and a pile of plugins, everything — pages, articles, forms, submissions, media, sections, saved queries, even your design theme — is a single kind of object connected by typed relations in one content graph. Publishing, knowledge graph, forms, media, and consent management are simply different views of the same store.
On top of that graph sits an agentic subsystem: twenty built-in AI jobs powered by leading large language models via the OpenRouter gateway (model-configurable). Agents propose relations, generate summaries and keywords, detect duplicates, draft complete pages, compose visual layouts, design entire themes — and even screenshot their own rendered output to critique the result. Nothing is published automatically: every recommendation lands in a review queue, and every run is logged with model, tokens, latency, and cost.
Despite the modern architecture, PressGraph keeps WordPress-style hosting ergonomics: PHP and MariaDB, no JavaScript build pipeline, no container orchestration required. PressGraph is developed by premotion media; availability and licensing are provided on request.
Features
Thought through to the daily workflow.
Every building block is focused on a clear benefit and fits into existing processes.
Feature Section
Every piece of content is an object; every connection is a first-class typed relation. Ten relation types ship out of the box — links, hierarchy, featuring, canonicals, translations, form submissions, query targeting, cover images, and embeds — all defined declaratively in configuration. Links written in the body are automatically synchronized into the relation store, so your content graph is always real, never an afterthought.
Graph-Native Content Model — One Store, Many Views
Every piece of content is an object; every connection is a first-class typed relation. Ten relation types ship out of the box — links, hierarchy, featuring, canonicals, translations, form submissions, query targeting, cover images, and embeds — all defined declaratively in configuration. Links written in the body are automatically synchronized into the relation store, so your content graph is always real, never an afterthought.
- One object store and one relation store power publishing, forms, media, and system views
- Ten typed relation types, declaratively configured
- Wikilink-style authoring with automatic body-to-graph synchronization
- Mandatory revisions on objects and relations — restore creates a new revision, never rewrites history
- Interactive graph visualization of any page's neighborhood, in the admin and on the public site
Agentic Subsystem — Twenty AI Jobs with Full Governance
The agent runtime covers content, media, design, and layout: propose relations, generate summaries and keywords, suggest cluster membership, detect duplicates, surface content gaps, draft complete pages, generate alt text from image pixels, design and adapt themes, compose page layouts, and more. Beyond one-shot generation, conversational refine lets you iterate on a draft with follow-up instructions while preserving full run lineage — and visual self-critique lets the layout agent screenshot its own rendered page and judge the pixels with a multimodal model.
- Twenty built-in agent jobs across content, media, design, layout, and templates
- Suggestion queue with explicit accept/reject — agents never publish on their own
- Unconditional audit log per run: prompt version, model, latency, token usage, and cost
- Conversational refine with recorded run lineage; visual self-critique of rendered layouts
- Disabled by default — the platform is fully usable without any AI calls
- Runnable from the admin UI or the command line, including safe bulk runs
Visual Layout Composer — Eighteen Block Types, Zero Build Step
Compose pages from rows, columns, and blocks on a twelve-column responsive grid. Eighteen block types ship, from markdown, media, forms, and dynamic query lists to hero and cover rows, CTAs, feature grids, stats, quotes, marquees, galleries, and reusable Symbols you define once and embed anywhere. Click-to-edit Presentation mode, autosave with draft and published perspectives, and soft-lock editing with optimistic concurrency keep teams productive without stepping on each other.
- Eighteen block types including CTA, feature grid, stats, gallery, and reusable Symbols
- Hero/cover rows with background media, overlays, parallax, and scroll-reveal motion
- Click-to-edit Presentation mode, autosave, and live preview
- Editable page-template hierarchy: front page, single, archive, search, and 404 roles
- Agents can regenerate a single block, a row, or compose an entire page layout — always as reviewable drafts
Design as Content — Themes You Can Version, Preview, and Delegate
The design system itself is a content object. Design tokens, helper CSS, and font imports compile into a single served stylesheet with proper caching, and every change is versioned through the same revision system as your content. Theme agents can draft a complete theme from a brief, adapt an existing one, or generate named style variations — including a token-flip dark mode.
- Theme console with live preview and full revision history
- One-click activation with transactional handover between themes
- Token-first CSS — no utility-class framework, no frontend build pipeline
- Dark mode via token flip; agent-draftable style variations
- Cross-document view transitions for a smooth, app-like public experience
Built for Search — Lexical, Vector, and Hybrid Retrieval
Search is not a plugin — it is a core capability with three retrieval axes. Full-text lexical search, embedding-based vector search, and a hybrid mode that fuses both via reciprocal rank fusion, available on the public site and in the admin.
- Full-text lexical search out of the box
- Embedding-based semantic vector search with a chunking and indexing pipeline
- Hybrid retrieval via reciprocal rank fusion
- Admin search variant that also surfaces drafts
Topical SEO Engine — Clusters, Link Health, and Clean Markup
PressGraph treats topical authority as a measurable structure, not a hope. Algorithmic clustering groups related pages and writes proposals to a review queue. Per-hub scorecards and an alignment report show how well each cluster holds together, while the link-health audit flags orphan pages, weakly connected content, and anchor-text monoculture.
- Algorithmic clustering (label propagation) with human-reviewed proposals
- Per-hub scorecards, cluster alignment report, and related-content navigation with previous/next paging
- Link-health audit: orphans, weak connections, anchor-text monoculture
- Multilingual routing with hreflang tags, XML sitemap, robots directives, and canonical redirects
- Breadcrumbs with structured data markup, heading permalinks, and a topical meta cascade from hubs to leaf pages
Lead-Gen Forms — Conversion Without the Spam
Build forms visually, embed them anywhere with a simple shortcode-style tag, and collect submissions as first-class objects linked back to their form. A layered anti-spoofing stack — honeypot, signed submission tokens, timing trap, and rate limiting — keeps the inbox clean without annoying real visitors.
- Visual form builder with text, email, textarea, checkbox, select, and hidden fields
- Honeypot, HMAC-signed tokens, timing trap, and configurable rate limits
- Submissions inbox in the admin; optional email notifications
- Submissions stored in the content graph, linked to their form
Privacy First — Consent, Events, and Auditability
A built-in consent banner manages essential, analytics, and marketing categories with a privacy-first default: essential only. The server-side event bus is consent-gated per listener, so analytics and marketing integrations only ever fire for visitors who agreed — and every event is persisted to an audit log.
- Consent banner with essential / analytics / marketing categories, essential-only by default
- Consent-gated server-side event bus with persistent audit trail
- Configurable IP storage policy for form submissions
- No frontend tracking scripts required by the platform itself
Extensible by Design — Plugins with Guard Rails
Drop a plugin folder with a manifest and it is discovered, validated, and wired into the event bus automatically. Plugins run against a capability-gated SDK — events, outbound HTTP with a host allowlist, read-only object access, and namespaced logging — so an integration can never quietly exceed the permissions it declared.
- Manifest-declared event subscriptions, auto-wired on load
- Capability-gated SDK: events, HTTP with host allowlist, read-only objects, logger
- Enable and disable plugins from the admin, with manifest inspection
- Example webhook plugin included: forward form submissions to any external system, consent-gated
How it works
A clear path to the result.
- 01
Step 1 — Install on Standard PHP Hosting
Deploy the codebase, point it at a MariaDB database, run the migration command, and set your admin credentials. No container platform, no Node.js build, no external services required to go live.
- 02
Step 2 — Author Content in the Graph
Create pages and articles in the HTMX-powered admin editor with markdown, wikilinks, live slug checks, and a local-graph sidebar. Compose visual layouts from eighteen block types, or let the body markdown render on its own.
- 03
Step 3 — Switch On the Agents (Optional)
Add your OpenRouter API key and enable agent jobs. Choose your preferred model — configurable separately for agent jobs and embeddings. AI stays off until you explicitly turn it on.
- 04
Step 4 — Review Every Suggestion
Agents write proposals to the suggestion queue: new relations, summaries, keywords, drafts, layouts, themes. Accept or reject each one. Accepted artifacts are written through the same services human edits use, with agent attribution in the revision history.
- 05
Step 5 — Measure and Improve
Watch cluster scorecards, the alignment report, and the link-health audit. Run content-gap analysis, refine drafts conversationally, and let the layout agent critique its own rendered output — while the audit log tracks every model call and its cost.
Who is it for?
Made for concrete requirements.
The solution is a particularly good fit for these teams, roles and business models.
Digital Agencies
Deliver content-rich client sites on infrastructure you already know how to host, with an AI production line built in. Every agent action is attributed and auditable — exactly what client governance demands.
SEO and Content Teams
Topical clusters, link-health audits, hybrid search, hreflang, sitemaps, and structured data are core platform features, not a plugin stack. Content-gap and keyword agents feed the editorial calendar.
Publishers and Knowledge-Driven Organizations
The graph model treats connections between content as first-class data. Build genuine knowledge structures — hubs, translations, canonical chains — and visualize them interactively.
Marketing Teams
Lead-gen forms with serious anti-spam, conversion-oriented layout blocks (CTAs, stats, feature grids, pricing-style compositions), and a consent-gated event pipeline for clean, compliant analytics.
Privacy-Conscious European Businesses
Privacy-first consent defaults, configurable IP storage, server-side events gated on visitor consent, and full audit trails align with European data-protection expectations.
Developers
A disciplined, modern PHP codebase: strict layering, hand-written SQL, declarative configuration for object types, relation types, and blocks, a capability-gated plugin SDK, and a full CLI.
Tech & scope
The key details at a glance.
- Product Type
- Self-hosted agentic content platform (CMS)
- Language / Runtime
- PHP 8.4
- Database
- MariaDB 10.6+
- Frontend
- Server-rendered Twig, HTMX + Alpine.js, no build pipeline
- AI Integration
- OpenRouter gateway; model freely configurable
- AI Default State
- Disabled until explicitly enabled
- Built-in Agent Jobs
- 20
- Block Types
- 18
- Relation Types
- 10
- Search
- Lexical, vector, and hybrid (reciprocal rank fusion)
- Multilingual
- Configurable languages, hreflang, translation relations
- Admin UI Language
- English
- Consent Management
- Built in (essential / analytics / marketing)
- Extensibility
- Manifest-based plugin runtime with capability-gated SDK
- Test Coverage
- 300+ automated tests plus static analysis
- Author
- premotion media
- Licensing & Availability
- On request from premotion media
FAQ
Frequently asked questions.
How is PressGraph different from WordPress?
WordPress stores posts and pages and delegates almost everything else to plugins. PressGraph stores every kind of content as objects in one graph with typed relations, and ships search, forms, media, consent, clustering, theming, and AI agents as core platform features. It deliberately keeps WordPress-style hosting ergonomics — PHP plus MariaDB on standard hosting — while replacing the internal model.
Do I have to use AI to use PressGraph?
No. The agentic subsystem is disabled by default, and the platform is a complete CMS without it: editor, layout composer, forms, media, search, themes, and multilingual publishing all work with AI switched off.
Which AI models does PressGraph use?
AI calls go through the OpenRouter gateway, giving you access to all major large language models. The model is freely configurable — including a separate model choice for agent jobs and for embeddings.
Can the AI publish or change content on its own?
No. Every agent output lands in a suggestion queue for explicit accept or reject. Accepted results are written through the same service layer as human edits, attributed to the agent in the revision history. Every run is logged with prompt version, model, token usage, latency, and cost.
What does "visual self-critique" mean?
The layout-critique agent renders the page it just composed, captures a screenshot, and feeds the image to a multimodal model to evaluate the actual visual result — not just the markup. Its findings come back as reviewable suggestions like everything else.
Does PressGraph support multilingual websites?
Yes. Supported languages are configurable, with language-prefixed routing, hreflang alternate tags in pages and the XML sitemap, and explicit translation relations connecting language versions of the same content.
What are the hosting requirements?
PHP 8.4 and MariaDB 10.6 or newer. There is no JavaScript build step and no container requirement, so the platform deploys on standard PHP hosting environments. Hosting and setup are available from premotion media on request.
How do I get PressGraph?
PressGraph is developed by premotion media. Licensing, deployment options, and pricing are available on request — contact premotion media for a demo and a consultation.
Ready?
A CMS That Works With You — Agentic, Auditable, and Built on a Graph
PressGraph combines a graph-native content model, a visual layout composer, hybrid search, topical SEO tooling, and twenty governed AI agents into one self-hosted platform with familiar PHP hosting ergonomics. The AI proposes; you decide; the audit log remembers. Developed by premotion media — licensing, hosting, and deployment available on request.
Discuss your project