how it works
The agent's pipeline.
Your resident agent runs every article through ten stages — from live research to a published, indexed, cross-linked post with a cover image and full metadata. About two minutes end to end, on a schedule, with nobody watching. Here's each stage, then the safety systems wrapped around them.
Intelligence Gathering
Pulls real-time signals from multiple sources — breaking news, trending repositories, industry shifts. Multi-region support so content matches your audience's market. Today's signals, not last month's training data.
Topic Selection
The system maintains a topic queue per site, filled by the weekly SEO Brain. If the queue has pending topics discovered from actual search data, it pulls the highest priority one. Otherwise, the LLM analyzes today's intelligence against your site's themes and picks an editorial angle. Built-in deduplication ensures nothing repeats.
Competitive Analysis
Searches the actual search engine results for the chosen topic. Pulls the top competing articles, People Also Ask questions, and related searches. The generation stage knows what's already ranking — and is told to write something better.
Context Assembly
Gathers everything the system knows about your site — featured content, brand context, site profile, audience signals, internal data. This context shapes the article so it reads like it was written by someone who knows your business.
Cross-Link Discovery
Queries the network's content registry for relevant articles from sibling sites. The matching engine scores every candidate by multiple relevance signals. Only the strongest matches make the cut.
Article Generation
The LLM writes a full long-form article with your site's voice — persona, tone, structure, and content rules enforced. The prompt includes competitive intelligence, cross-links, site context, SEO targets, and search intent signals. The system doesn't generate content. It generates content that outranks.
Structure & Navigation
The article is post-processed for structure — table of contents, heading hierarchy, internal anchor links, reading flow optimization. Ensures the piece works as a standalone resource, not just a wall of text.
Cover Image
Generates a branded cover image with your site's visual identity — colors, logo, typography. Optimized for social sharing and open graph previews. Each site gets a distinct visual style so no two network sites look alike.
Publish & Index
Writes the article to your database with full metadata — title, slug, excerpt, body, tags, category, SEO fields, cover image. Updates the network's content registry so future articles across the network can discover and link to this piece.
Network Propagation
The content registry updates propagate across the network. Existing articles in sibling sites become eligible for cross-linking to this new piece. The network's authority graph rebalances. Every new article makes the entire network slightly stronger.
What runs in the background
The pipeline is just the visible part. Behind it, four autonomous systems run continuously to keep the network healthy, the content strategy sharp, and the authority growing.
SEO Brain
Runs weekly. Researches actual search engine data for every theme on every site. Discovers keyword gaps, trending questions, and content opportunities. Generates topic ideas and queues them for the pipeline. Your content strategy evolves with the search landscape.
Orchestrator
Runs daily. Self-monitoring system that checks pipeline health, site uptime, content velocity, keyword rankings, topic queue depth, and network health. Generates actionable recommendations and flags issues before they become problems.
Network Safety
Runs continuously. Monitors every site in the network for spam, scam patterns, toxic backlinks, and SEO violations. Flagged sites are disconnected and dropped from the content registry, so nothing in the network links to them again. The network protects itself.
Weekly Digest
Every Monday morning. Articles published, keyword rank movements with direction arrows, topic queue status, pipeline success rate, and recommendations. Delivered to your inbox. No login needed. No dashboard to check. Just the information you need.
The gates — why autopilot doesn't mean slop
Autonomous publishing has a known failure mode: low-quality content at scale. The pipeline is built around refusing it.
Quality gate
Every draft faces an LLM judge checking for hollow claims, missing evidence, and structural failures. Blocked drafts get one targeted revision against the judge's exact verdicts, then re-face the full gate. Still failing? The slot is skipped — no post is better than a bad post.
Approval policies
Per-site and per-action-class: approve each item from your inbox, hold for a grace window that auto-publishes if you're busy, or run full auto. The gate and the policy are independent — auto skips your approval, never the quality check.
Disclosure & lineage
Every article carries an AI-disclosure footer and a public lineage page showing how it was made — sources, checks, and the decisions in between. Trust is a feature of the category.
The capability ladder — how much it can touch
The agent starts with the narrowest possible grant and earns access. You hold the keyring — revoking a connection removes the capability instantly.
Any website
Content, SEO, indexing, distribution, tracking, reports. Works with your existing database or hosted mode. This is where every site starts.
Pages we serve
On hosted or proxy delivery, the agent can create and update landing pages on surfaces it already serves — instantly, no rebuilds.
Your repositoryrolling out
Grant the GitHub App and the agent proposes changes as pull requests — performance fixes, metadata, new pages. Nothing merges without your review.
Your CMSrolling out
WordPress, Shopify, Webflow adapters — the agent operates the platform you already publish with.
Build News mode (optional)
If you're building a product in public, Networkr can turn your real shipped work into weekly build-log posts — first person, verifiable, indexed as you build. Two ways to wire it up. Both use the same pipeline, the same quality gates, and the same immutable archive. Pick the one that matches how much control you want.
Fully automated
Connect GitHub once. Our pipeline pulls a week of commit metadata — SHAs, messages, dates, changed file paths — clusters by theme, extracts decisions from PR titles and ADRs, and writes the post. You do nothing.
- ✓Read-only scope. Metadata only — never code, never diffs, never file contents.
- ✓Works on private repos.
- ✓/private: in a commit message excludes that commit entirely.
- ✓Metadata is purged after the post publishes. Only the finished article persists.
Your agent via API
No OAuth. npx networkr init drops .networkr/AGENT.md and a post-merge hook in your repo. Your own IDE agent reads git log locally, filters, clusters, and POSTs a structured narrative to our API. Only the narrative crosses the wire.
- ✓Code never leaves your machine — architecturally impossible. The payload schema has no field for diffs.
- ✓Hook fires on merge, or run npx networkr report manually whenever.
- !Semi-automated. If your agent isn't available when the hook fires, no post gets generated that week.
- !Silence on your side = silence on the blog. We never fabricate a post to hit cadence.
Both paths converge on the same pipeline, same prose guardrails, same lineage record, same immutable archive. Path A trades control for zero touch. Path B trades zero touch for full data sovereignty.
Voice System
Every site has a voice that controls how articles are written. Voice is separate from author — voice is the style, author is the byline. The system enforces voice rules at generation time, including banned words, structure preferences, and tone constraints.
Custom voices supported — define your own persona, tone, banned words, structure rules, and preferred model.