The modern AI toolstack for serious LinkedIn creators

The modern AI toolstack for a serious LinkedIn creator is a short list of jobs, not a logo wall. You need a place to hold a source you can stand behind, a model call that writes from that source, a way to turn the same source into a 2–8 slide PDF, a way to talk when typing is slower, a feed check, and a clock that publishes one personal profile. Everything else is optional until one of those jobs is failing in public.

Boykli is that list on one seat. Create posts writes from a note, a YouTube URL, an article URL, a document, or a series source. Voice Agent turns a live interview into three drafts from the transcript. The carousel page builds the PDF with the brand kit on the same screen. Schedule publishes the personal profile you connected, on the weekdays and the one time of day you saved. Library keeps the drafts. Radar is a reading list of public URLs. Algorithm is a plain-language read of LinkedIn's public notes on distribution. The product does not comment, like, connect, or message.

Logged-in visitors can browse Radar and Algorithm with no plan. Writing, carousels, images, Voice Agent, Post to LinkedIn, and Add to Queue need a paid seat. BYOK is $29 a month: your OpenRouter key for chat and images, and a Google AI Studio key for Voice Agent and for YouTube when the video has no captions. Boykli does not add a monthly generation cap on that seat. Managed is $88 a month and includes the platform models inside caps: 210 posts across the seven write modes, 150 carousels, 150 images, 25 Voice Agent sessions, and 130 post refinements. On Managed, a single draft allows two rewrites. Prices are the public list on boykli.com as of 27 October 2026.

What a serious stack is for

People who search this phrase have already tried a general chat tab. The draft came back shaped like a topic prompt: three lessons, a round percentage, a closer that asks for comments. They then opened a design tool for a PDF, a formatter for bold letters, a scheduler, and an analytics extension, and the post still died in the handoff. The search is for the minimum set that finishes a post a stranger can use.

Serious, in this stack, means four constraints:

1. The source is yours. A decision, a number you measured, a sentence you are allowed to paraphrase, a URL, a file, or something you can say out loud.
2. The model is told to keep the nouns in that source and to drop stats, studies, quotes, and anecdotes that are not in it.
3. You read the draft before it can leave. A statistic you did not supply gets deleted.
4. Publish goes through LinkedIn OAuth on one personal profile (urn:li:person), now or on the next open slot. Company Pages are a different product.

A stack that starts from someone else's viral post, or that likes and comments while you sleep, fails constraint 1 and constraint 4. The reason this product starts from something you brought is the decision in building Boykli. The hour that runs one post through the list is the LinkedIn content creator workflow. This page is the tool map that workflow assumes.

Which layer does which job

The scattered column can still produce a good week if you are strict about the handoff. Most weeks the handoff is the work: the chat draft, the PDF, and the caption live in three accounts, and the document post has no clock. The composer is enough for one post you are ready to type today. It is a weak system for a weekday cadence. The Boykli column shares a draft: you cannot queue a post that does not exist yet, and a carousel is a separate artifact with its own caption, not a second paste of the same text.

Free LinkedIn Utilities: Turn plain text into clean 1080x1350 multi-page PDF slides using the Free LinkedIn Carousel Generator, and test your post hook and dwell score on the Free LinkedIn Algorithm Grader.

The layers, in the order you actually use them

1. A source, before any model

Open a note. Write the point in one sentence, then the proof you have. A number from the product, a ticket you are allowed to paraphrase, the date of a decision, the mechanism. If the source is a public URL, save it in Radar or paste it into article mode. Article mode writes from the page the fetcher can read. A page the fetcher cannot read does not become a confident summary.

Radar is the reading layer of the stack. It is not a swipe file of posts to rewrite, and it is not a score of which creator won overnight. Algorithm, on a different morning, explains how LinkedIn describes distribution. It is not a recommendation of what to post tomorrow.

Empty capture is the failure that no model fixes. "Thought leadership about onboarding" is a label. The prompts are written to use what you supply. A label still produces a thin post, because there was nothing to keep.

2. One draft, in a length the feed can finish

Pick one of the seven modes. The choice is the shape.

• Short, medium, or long when the source is a note. Medium is the default for a decision with two numbers and a reason. LinkedIn's own ceiling is 3,000 characters.
• YouTube when the source is a watch URL. The draft is a teaser of the video. If the video has no captions, the watch uses Google AI Studio on BYOK and Vertex on Managed. A failed watch returns an error. It does not invent a post from the title.
• Article when the source is a public page.
• Document when the source is a file you already wrote.
• Series when one source needs 2 to 10 posts. Each part is its own slot later, up to 8,000 characters. A series of four on a three-day week consumes more than a week.

Posts default to Grok 4.6. On Managed that call is xAI's Responses API with web search. On BYOK it goes through OpenRouter with the web plugin. Create posts has no model picker. If OpenRouter reports that Grok 4.6 is at capacity, the post page tells you which model was used for that draft. That notice is a capacity fallback. It is not a license to send an arbitrary model slug from the browser. How a draft is shaped once a key is present is the LinkedIn post generator.

Optional controls sit next to the draft, not in other products. Alternate hooks propose first lines from the same note. A call to action is a line you fill when there is a real next step, and a switch you turn off when there is not. An image, if the post needs one, uses one of the four styles above. You still attach it to this draft. There is no separate image studio to manage.

On Managed, two rewrites per draft count toward the 130 refinement cap. BYOK does not apply that Boykli rewrite ceiling. The provider invoice is the ceiling. A rewrite is still a new model call. It is not a free pass to keep a sentence you would not say.

3. A document, only when the source is a sequence

A carousel is a PDF LinkedIn paginates. The first page is the card in the feed. The caption is a separate text, up to 3,000 characters, and the document title is required when you queue the file, capped at 120 characters. Square is 1080×1080. Portrait is 1080×1350. Slide count is 2 to 8.

The design list on the carousel page is Neo-Brutalist (the default), Bauhaus, Retro Terminal, Infographic / Data Guide, Bold & Edgy, Creator Gradient, Split-Screen, and Normal. Normal is the layout that uses the brand kit colors. The brand kit itself — canvas, accent, logo, avatar, handle — lives on the carousel page, not in Settings. The model picker for carousel copy is Grok 4.6, Gemini 3.5 Flash, Claude Sonnet 5, or gpt-5.6-terra. Anything off that list is rejected. On Managed, Gemini goes through Vertex. On BYOK, the same four names go out on your OpenRouter key.

Build the PDF after the text post is queued, from the same note, when the note is a sequence someone can swipe. It takes the following slot. It is not a second caption of the same draft. The full generator path is LinkedIn carousel generator. Scheduling a file that already exists does not call the model again.

Skip the PDF on a short day. A text post that contains the point is a finished artifact. A deck with no point is a file.

4. Voice, when the source is spoken

Voice Agent is the layer for a source you can say faster than you can type. You talk. The session is transcribed. Three drafts come back from that transcript, not from a topic label you typed in a chat box.

On BYOK the live session uses your Google AI Studio key only. It is never rewritten onto Vertex. On Managed the live session uses Vertex. Drafting after the interview uses OpenRouter on BYOK, including when the interview itself already ran on Google, and platform models on Managed. A missing Google key stops the live interview. The screen says which key is missing and links to Settings. It does not complete the call on Boykli's credentials.

Create posts is still the better layer when you already know the shape and you have a sentence written down. Voice Agent does not ask you to pick a length first. The interview, the transcript, and the three drafts are covered in LinkedIn Voice Agent.

5. A clock on one personal profile

Add to Queue takes the next open slot from the days, the one time of day, and the timezone you saved. You do not type a calendar date into the queue. Drag changes which draft gets which date. Posts ahead and Scheduled through are the check that the row landed. A new queue starts on Monday, Wednesday, and Friday at 10:00, in the timezone of the device, until you click Save. Change that once.

Post to LinkedIn is for now. Use it when the post is late for the slot you care about. Putting five drafts on that button in one sitting publishes five times today.

The queue skips a send when the LinkedIn token is unusable, when no weekday is selected, when the timezone was never saved, or when the account no longer has a generatable plan. A cancelled seat with current_period_end still in the future can still publish. on_hold and an expired cancel cannot. If the header says Reconnect LinkedIn, the rows stay and nothing goes out until you reconnect. The queue itself is LinkedIn scheduling tool.

Company Page posts stay outside this stack. The connection is the member profile, scope including w_member_social. A Page tool has a different clock and a different admin role.

Assemble it in one afternoon

Do this once. After that, the weekday is the workflow page, not a fresh tool decision.

1. Create the seat. BYOK at $29 a month if you already pay OpenRouter and you want Google AI Studio for voice. Managed at $88 a month if you want platform keys and you can live inside 210 posts, 150 carousels, 150 images, 25 Voice Agent sessions, and 130 refinements. An AppSumo lifetime code, redeemed at boykli.com/redeem, is also a BYOK seat: one code, one account, and an existing paid seat blocks redeem. A later Managed checkout can still start. Prices and caps are the public cards. Recheck them on boykli.com before you quote them to a team.
2. Save the keys if the seat is BYOK. Settings → LLM API Keys. OpenRouter keys come from openrouter.ai/keys. Google AI Studio keys come from aistudio.google.com/app/apikey. A string that does not look like an OpenRouter key is rejected, and a key OpenRouter itself rejects is not kept. Keys are stored encrypted and are not shown back in full. Managed has no user key on this path.
3. Connect LinkedIn. One personal profile. Confirm the header is that profile, not a Page you admin. Disconnect is available when you want the token gone.
4. Set the brand kit before the first PDF. Canvas, accent, logo, handle. Do this on the carousel page so the first document is not a default skin you then redo.
5. Save the clock. Pick the weekdays you will actually ship, one time of day, and the timezone. Look at Posts ahead before you write, so you do not draft into a slot that is already taken.
6. Write one real post from a note you would sign. Use medium unless the note is a single fact (short) or a full objection with a number (long). Read it out loud. Delete any study, name, or percentage that was not in the note. Queue it. Stop.
7. Add a second artifact only if step 6 was a sequence. Same note, carousel, next slot. Read the caption as feed text. Do not paste every slide into the caption.
8. Close the other tabs. If the draft, the PDF, and the queue all landed here, the chat tab, the design export, and the third-party scheduler are no longer on the critical path. Keep LinkedIn's own analytics if you want impression counts. This stack does not replace that screen.

A worked source, so the afternoon is concrete:

Tuesday we cut the shared free pool. It was 1,000 API calls a day for the whole account. It is now 250 calls a day per workspace. The ticket we paraphrased, with permission: one script was starving the other three apps. We count successful responses, not page views.

That note is enough for a medium post today and, if you want the mechanism as slides, a short portrait PDF on the next slot. It is not enough for a series of eight. A series needs more than one decision.

What to leave out of the stack

Some tools show up in every "AI stack for creators" diagram and do not belong next to a serious publishing loop.

• Engagement pods, comment bots, and connection automations. They act on LinkedIn as you, outside the post you approved. Accounts get restricted for that class of automation. Boykli's send is the post or the document you queued, through OAuth, one due item at a time.
• A second scheduler. Two clocks on one profile produce double posts or a draft that exists in a product you forgot to pause. Pick the Boykli queue or LinkedIn's native clock for a single post. Do not run both against the same week.
• A font generator that restyles the whole body. LinkedIn bold and italic are a Unicode letter swap. A screen reader can lose the sentence. One plain-text label is the ceiling. A line break you can scan is the layout.
• An analytics suite inside the writing tool. Impressions live on LinkedIn. A writing stack that also promises to pick your next topic from last week's likes will chase the number and drop the source. Read LinkedIn's numbers on LinkedIn.
• A company Page scheduler, if the audience is your name. The personal profile is the account this OAuth can post as. Forcing Page drafts into this queue will not make them Page posts.

If a tool's only job is one cell of the table above, and that cell is already on the seat, cancel it after the first week you shipped without opening it. Subscriptions you do not open are not a stack. They are a bill.

FAQ

Do I still need ChatGPT or Claude if I use Boykli?

For the LinkedIn draft, the PDF copy, the image, and the voice interview, no. Those calls run on the seat: your OpenRouter and Google keys on BYOK, platform keys on Managed. Keep a general chat tab for work that is not a LinkedIn post. Pasting a topic into that tab and then into LinkedIn is the workflow this stack replaces. Claude, when you do use it for a document, is the carousel picker name Claude Sonnet 5, not a separate writing product you have to export from.

Is Taplio, Buffer, or AuthoredUp part of this stack?

Only if you still want their specific job after the Boykli seat is doing it. Taplio and Buffer are queues, often with a company-Page bias and an AI allotment the vendor meters. AuthoredUp is a formatting and preview layer in the browser. Boykli's queue, preview read, and draft are the same seat as the model call. Recheck those vendors' pricing pages before you compare invoices. The Boykli prices to compare against are $29 and $88 a month, plus provider usage on BYOK.

Which seat should a working creator buy?

BYOK if you already have an OpenRouter habit and you want no Boykli generation cap. Budget the provider bill on top of $29. Managed if you want one invoice and your month fits inside 210 posts, 150 carousels, 150 images, 25 voice sessions, and 130 refinements. Both seats open the same screens. The plan row decides who pays the model. A missing BYOK key fails the call. It does not borrow a platform key.

Can this stack post to a company Page, comment for me, or message leads?

No. Publish and the queue target the connected personal profile. Comments, likes, invitations, and messages stay in LinkedIn, typed by you. A tool that offers those as "the AI stack" is a different risk category.

How many tools should be open during the writing block?

Three surfaces, in order: the note, Create posts or Voice Agent, then Schedule. Open the carousel page only after the text post is queued and the source is a sequence. Radar is for a different ten minutes, when you are saving a URL, not when you are trying to finish a sentence. If a fifth tab is open, the stack has already failed.