01
Compose the pipeline
Drag blocks onto a canvas and wire them together. An idea, a URL, a web search, or your last twenty posts go in one side; posts, articles, newsletters, images and hashtags come out the other.
Visual AI pipelines for LinkedIn
One idea in.
A week of posts out.
Compose a pipeline once — research, drafting, styling, a quality gate — then run it on one topic or twenty. Every draft is checked against your rubric and written in your voice before you ever read it.
No credit card. Bring your own model key — you pay the provider at cost.
Four stages, one canvas
Most AI writing tools are a text box with a send button. This is a graph you build once and run every week.
02
Run it on real input
Every output block runs in dependency order, reusing what upstream produced. Test a single node while you build, or run the whole graph. Each run is recorded with its prompt, its cost and its timing.
03
Gate it before it ships
Deterministic checks run first — length, hook, hashtag count, invented figures. Then an LLM judge scores against your rubric. A strict failure caps the score no matter how well the rest reads.
04
Feed the winners back
Record what a published post actually did. The next run ranks your best performers first and uses them as the examples it writes toward. The pipeline gets better because your results get fed back into it.
Built to be lived in
Pan, zoom and drag stay at sixty frames per second with a graph full of generated content. Edges follow a node while you move it. Every destructive action is one undo away.
Undo that means it
Every structural edit is reversible, and undo never throws away a generation you already paid for.
Sixty frames, loaded
Dragging a node moves it on the compositor and redraws only its edges. Nothing re-renders.
Keyboard first
Marquee select, group drag, snap to grid, duplicate, nudge, zoom to fit. Press ? for the full set.
Nine kinds of node
Each one does a single job and hands its output to whatever you wire it into. Ports are typed, so the canvas only offers you connections that make sense.
Post group
Post body plus hooks, CTA, hashtags and an image in one run.
Article
Long-form, sectioned, with inline images and a cover.
Newsletter
Subject line, preview text, banner and body sections.
Multiplier
Fan one input into N variants to pick between.
Evaluator
Score against your rubric. Strict criteria cap the score.
Web research
Search, date-filter, and feed sourced context downstream.
Production context
Your published posts as voice, anti-repetition or top performers.
Post image
Generated art directed by your saved image style.
Hashtags
A relevant set, derived from the finished post.
It writes toward what
actually worked
A generator that never sees its own results is guessing forever. Record what a published post did, and the next run ranks your winners first and writes toward them — so the fourth week is measurably better than the first.
- Publish
Post it, wherever you post. Nothing here needs access to your account.
- Record
Type four numbers off LinkedIn analytics — impressions, reactions, comments, reposts.
- Rank
Engagement rate, not raw reach, orders your library. A small post that landed beats a big one that did not.
- Rewrite
Your top performers become the few-shot examples the next run writes toward.
Then it runs again — with your own best posts as the target.
A week of posts in one pass
Switch any text input to batch mode and paste one topic per line. The pipeline runs once per line, sequentially, and you pick through the results afterwards — up to twenty-five in a run, stoppable at any point.
Nothing unverified goes out
Deterministic checks first
Length against the LinkedIn limit, hook strength, hashtag and emoji counts, sentence length, question ratio. Facts, not opinions — and they run before any model is asked.
Figures traced to source
Every number in a generated post is checked against the source it came from. A statistic that is not in the source is flagged as unverified before you publish it.
Self-repetition detection
New drafts are compared against what you have already published, so the same opening line does not quietly reappear every third week.
Roles that hold
Owners, editors and members. Members can run a published pipeline but cannot rewire it. The canvas goes read-only rather than pretending.
Your keys, your content
Generations run against your own model key. Nothing is used to train anything, and what you produce stays in your workspace.
Every run recorded
The compiled prompt, the model, the duration and the cost for each step. When a post reads badly you can see exactly what produced it.
Before you sign up
What is LinkedFlow?
A visual pipeline builder for LinkedIn content. You compose a graph of blocks — inputs like an idea, a URL or a web search, and outputs like a post, an article, a newsletter, an image or hashtags — then run it. The same pipeline can be re-run every week, or run over a list of topics in one pass.
How is it different from asking ChatGPT to write a post?
A chat window starts from nothing each time. A pipeline is composed once and keeps its structure: the same style, the same CTA resources, the same quality rubric, the same voice examples from your published history. It also records every run, checks figures against their source, and scores drafts against your criteria before you see them.
Does it post to LinkedIn automatically?
Not today. LinkedFlow takes an idea all the way to finished, checked, on-brand content and stores it in your library. Publishing is still your call, and you copy the result out when you are ready.
Can I use my own writing style?
Yes. The Style Library holds reusable post and image styles, and you can build one from examples of your own writing — the tool extracts the structure, tone and opening patterns into a style sheet you can then edit by hand.
How does it know what my best posts look like?
You record the real numbers on a published post — impressions, reactions, comments, reposts. Engagement rate is derived from those, and the Production Context block can rank your library by it, so the strongest posts become the examples the next generation writes toward.
Can my team use the same pipelines?
Yes. Invite people as editors or members. Editors can build and change pipelines; members can run published ones but not rewire them. Everyone shares the same style library, CTA resources and content library.
What does it cost to run?
Generations run against your own model API key, so you pay the provider directly at cost. Every run records its token usage and an estimated price, and the Run Inspector shows the figure per step.
What happens to the content I generate?
It stays in your workspace. Saved items keep full version history, so an edit or an AI variation never overwrites what came before, and generated images are re-hosted rather than left on a temporary CDN URL that expires.