The nodes, one by one
Every node in the workflow palette: what it does, what it needs before it, what it hands on after it, and what it costs.
Two kinds of things flow through a workflow
A workflow moves items from node to node, and an item is one of two things: a person (a contact or a search result) or a piece of content (a post, an article, a feed entry). Every node expects one of the two, and the canvas refuses a wire that hands it the other — the red warning names the node to use instead. Keep this in mind and most chains build themselves:
- People come from Profile List and People Search. They feed Visit Profile, Send Connection, Send Message, Enrich Profile, Withdraw Connection, the person conditions (Has Replied?, Is Connected?, Has Tag?, Already Visited?, Profile Enriched?, Reply Intent?) and Add Tag.
- Posts come from Discover Feed, Account posts, Your targets' posts, Posts on a topic, New followers (people who follow you on X) and from Publish Own Post when it publishes as a Page or on X. They feed React to a post, Comment on a post, Repost, Save post, Follow and Post Type?.
- Raw content comes from Content Sources, Fetch RSS feed, Fetch HTTP/JSON, Fetch Reddit, Fetch 4chan and Web search. It goes through Cluster topics, AI prompt, Languages, Quality check and Filter, and ends in Publish Own Post, Write an article or Write file.
What a node costs
Reading and writing on LinkedIn or X counts against the daily budgets of the channel (see Safety). Calling the AI costs money on the managed tier. The nodes that call the model are AI prompt (once per item), Quality check (at most once per item, only for the faithfulness check), Publish Own Post when it rewrites the text for the channel, Comment on a post and Repost when the comment is "on the post's claims" (three calls), Cringe mode, Cluster topics (once per run) and the "Use AI" switch on Send Connection, Send Message and the contact comment. Everything else is free.
Triggers: where a workflow starts
A trigger has no input. It produces the first items of the run.
- Profile List — hands the people you pick to the rest of the chain. Parameter: the profiles. Use it for a hand-picked list: the ten prospects you met at a fair, the five people to follow up this week. Goes before any person node.
- People Search — runs a LinkedIn people search and hands over the results. Parameters: the query, the source (a live LinkedIn search or only what is already in your local database), max results. Live searches cost reading budget; keep max results small. LinkedIn only.
- Cron schedule — the clock. Parameter: when it runs, as a cron expression, with a plain-language preview. It only fires once "Run on its own" is switched on at the top of the workflow. Goes before Content Sources, Account posts or any chain you want to repeat.
- Discover Feed — reads your own feed and hands over the posts that match your interests. Parameters: channel, max posts, only posts never handled before. It emits posts, so what follows is React to a post, Comment on a post or Repost. Pro.
- Content Sources — reads the sources you configured in Content → Sources and hands over their entries. Parameters: source types, macro-area (a beat: only sources tagged with it), max items per source, bundle all into one item, fetch full text, only new items. This is the start of every content pipeline: Cluster topics and AI prompt come after. Pro.
- Account posts — reads the latest posts of one LinkedIn Page, X account or Threads account. Parameters: channel, the Page URL or the @handle, max posts, only posts never seen before. It is the "watch a Page" trigger: React, Comment, Repost or Follow come after.
- Your targets' posts — reads the recent posts of the people in Strategy → Target, a few people per run in rotation. Parameters: who to watch (active targets, also the connected ones, or the feed targets), people to read per run, max posts, ignore posts older than N days, include reposts, only new. Every person read is one reading budget unit. Feed it into AI prompt → Comment on a post to comment on your targets' posts. Pro, LinkedIn only.
- New followers — the people who started following you on X in the last N days. Parameters: channel, days, skip people you already follow, max people. Goes before Follow (follow back) or Visit Profile. X only.
- Posts on a topic — searches X for posts on a topic. Parameters: channel, the words to search, max posts, only posts younger than N hours, minimum likes, only new. Feed it into React, Comment (reply) or Save post. X only.
Actions: what Booost does for you
Actions take one item and act on it, then pass it on.
- Visit Profile — opens the person's profile and reads it for a few seconds, like you would. Parameters: channel, the account to visit (leave empty in a chain: it visits the person flowing through, or the author of each post found by the previous node), read time. A visit is often the first touch before a connection request.
- Send Connection — sends a connection request, with an optional note. Parameters: the note (supports {{variables}} from the templates), Use AI to personalise it. Needs a person upstream. LinkedIn only.
- Send Message — sends a direct message to a connected person. Parameters: the message template, Use AI. The delivery identifies the recipient by profile URL, so it never hits the wrong namesake. Needs a person upstream. LinkedIn only.
- Enrich Profile — reads the full profile through LinkedIn's own API (role, company, headline, follower count) without visiting the page. No parameters. Put it before a Filter or an AI prompt that needs the person's details. LinkedIn only.
- Withdraw Connection — withdraws a request that was never accepted. Parameter: after N days (enforce it with a Wait before this node). LinkedIn only.
- React to a post — leaves a reaction. Parameters: channel, Which post (the post from the previous node, or a recent post of the contact's profile), post index (for the contact case: 0 = the most recent), post URL (leave empty in a chain), react as (you or one of your Pages), the reaction (Like, Celebrate, Support, Love, Insightful, Funny). With "the post from the previous node" it needs a post upstream; with "a recent post of the contact's profile" it needs a person upstream. It passes the post on, so Repost can follow.
- Save post — bookmarks the post on X or Threads. Parameters: channel, post URL (empty in a chain). Needs a post upstream.
- Comment on a post — writes a comment. Parameters: channel, Which post (as above), when it runs (automatically, or ask me first in Content), what the comment should say (something of your own, or on the post's claims: they are checked on the web first), post URL, comment text and Use AI (for the contact case), post index, comment as (you or a Page). With the post from the previous node the text comes from an AI prompt before this node; with the contact's post the text is the template here. Pro.
- Follow — follows (or unfollows) a Page or an account. Parameters: channel, follow as (which of your Pages follows: LinkedIn has no person-to-Page follow here), the Page or account to follow (empty in a chain: the author of the post flowing through), action (follow or unfollow). Counts against the daily budget.
- Write an article (draft) — creates a LinkedIn article draft from title and body. Parameters: title and body (empty = what AI prompt put in the item), write as (you or a Page). It stops at the draft: you publish from LinkedIn. LinkedIn only.
- Publish Own Post — publishes a post, or saves it for your approval. Parameters: channel, text (empty = what arrives from AI prompt), Fit the text to this channel: only if over the limit (rewrites when the text does not fit — 280 characters on X), always (writes the post from what arrives, with your instructions), never (stops if too long); mode (save for approval, save with the quality gate, publish live); publish as (you or a Page); closing line (added word for word, with {{link}}). When it publishes as a Page or on X it passes the new post on, so React to a post and Repost can act on it.
- Repost — reshares a post. Parameters: channel, post URL (empty in a chain), repost as, with or without a comment, the comment text, what the comment should say, when it runs. Without a comment it is a plain amplification; with one it is a post of yours and counts against your daily post budget. Needs a post upstream.
Conditions: two exits
A condition takes one item and sends it out of exit 1 (yes) or exit 2 (no). Wire both, or the "no" side is dropped silently.
- Has Replied? — did the person answer your last message within N days? Parameter: within days. Person upstream.
- Reply Intent? — did the reply fall into one of these triage categories (warm, interest, hot, cold, spam)? Parameter: the categories to match. The category comes from the Inbox triage. Person upstream. Pro.
- Is Connected? — is the person a 1st-degree connection now? No parameters. LinkedIn only.
- Has Tag? — does the person carry this tag? Parameter: the tag. Pair it with Add Tag to build state into a chain ("tagged: contacted").
- Already Visited? — did Booost visit this profile within N days? Parameter: within days. Stops you from visiting the same person twice.
- Profile Enriched? — was the profile enriched already? No parameters. Put it before Enrich Profile to skip the ones already done.
- Post Type? — does the post match: allowed types (text, image, video, link, repost), must contain one of these keywords, must not contain these. Post upstream. The filter of every "watch and react" chain.
Wait
- Wait — pauses the item. Parameters: days, hours, and Up to (days): leave 0 for an exact wait, set it above Days to wait a random time between the two — less mechanical than the same delay every time. A chain "Send Connection → Wait 3 up to 5 days → Is Connected? → Send Message" reads like a person, not a script.
Utilities: shaping content and controlling the flow
- Quality check — three checks on the text, in order of cost, stopping at the first that fails: epistemic checks (seven local rules: absolute claims, missing hedges — free), cringe detector (one call to the Booost server, managed tier; skipped and declared when unavailable), faithfulness to sources (asks the model what share of the post's claims the source material supports, with the same subject, purpose and cause — the only AI call, last, only if the others passed). With Repair on (the default), a text whose claims the sources do not support is rewritten by the model without those claims and checked once more: up to two extra AI calls, only when needed, and no disclaimer is ever added. Only a text that still fails goes to exit 2. Parameters: the three switches, cringe threshold, minimum grounding score, repair, model override, reasoning effort. Exit 1 = passed, exit 2 = stopped, with scores and reasons on the item and in the run. Pro.
- A/B Split — sends a share of the items out of exit 1 and the rest out of exit 2. Parameter: variant A percentage. Two connection notes, two message templates, and the Analytics tell you which wins.
- Add Tag — adds or removes a tag on the person. Parameters: tag, action. It is how a workflow leaves a mark that another workflow or a Filter can read.
- Filter — keeps the items that match your rules, sends the others out of exit 2. Parameters: logic (all rules or any rule), the rules (field, comparison, value) on the person or on the content — company, role, location, headline, tags; title, text, link, topic, author, source name.
- Write file — writes the item to a file under the app's output folder. Parameter: file name. For reports and exports made by a workflow.
- Naturalness score (timing) — scores how human the timing of your recent actions looks, from timestamps a Database query put on the item. No parameters. An audit tool, not a content gate. Pro.
- AI prompt — the writer. Parameters: generation mode (from the next node — a post if it feeds Publish Own Post, a comment if it feeds a comment node — or comment, message, post, reply, purposes), output language, target length (recommended = 2,000–2,500 characters, where engagement peaks), system prompt, user prompt with {{variables}} from the item (title, text, link, mergedSources…), max tokens, model override, reasoning effort. It writes the result into the item's text, which the next node reads. One AI call per item.
- Database query — runs one of the registered safe queries (recent actions today, weekly engagement count, inbox unread count, recent engagement timestamps…) and writes the result into the item. Parameters: query name, the field to write to. Feeds Naturalness score and Detect anomalies. Pro.
- Detect anomalies — flags unusual patterns in your posting and engagement timestamps. No parameters. Pro.
- Fetch RSS feed — reads one feed. Parameters: feed URL, max entries, only articles never seen before, fetch the full article text. A single-source alternative to Content Sources. Pro.
- Fetch Reddit — reads a subreddit listing or one thread with its comments. Parameters: mode, subreddit, sort, thread id, max posts, comments per post, delay between requests. Reddit refuses some networks without login: if it fails, that is why. Pro.
- Fetch HTTP/JSON — reads any JSON endpoint and turns its records into items. Parameters: URL, array path, the title, link and text fields, max items. Pro.
- Fetch 4chan — reads a board catalog or one thread. Parameters: board, thread id, max threads, max comments, delay. Pro.
- Web search — searches the web and hands over the results as items. Parameters: query with {{variables}}, max results. The research step before an AI prompt. Pro.
- Cluster topics — groups the incoming entries by topic and merges the members, so two sources on the same subject become one item with the merged material and a title. Parameters: max topics, characters per source, max sources per topic, max input characters, fetch the full article of the merged sources (only for the sources that become a post, at most topics × sources per topic; a page that cannot be fetched keeps its teaser), minimum relevance, model, reasoning effort. One AI call per run, whatever the number of entries. Pro.
- Cringe mode (parody post) — writes a deliberately over-the-top post on a topic, for parody. Parameters: topic, max tokens. One AI call per item. Pro.
- Languages — multiplies the item: one copy per language, so one AI prompt after it writes the post in each. Parameter: the languages (empty = your profile's). Every AI node after it costs once per language. Pro.
How they combine: the chains that work
- Outreach to a list: Profile List → Visit Profile → Wait 1–2 days → Send Connection → Wait 3 up to 5 days → Is Connected? → (yes) Send Message. Add "Already Visited?" before the visit to skip repeats.
- Follow-up on replies: Profile List → Has Replied? → (no) Send Message. Or Reply Intent? to answer only the warm ones.
- Watch a Page and engage: Cron → Account posts → Post Type? → React to a post → Repost. Add AI prompt → Comment on a post to comment instead of reacting.
- Comment on your targets' posts: Cron → Your targets' posts → Post Type? → AI prompt → Comment on a post (ask me first). The comment waits for you in Content.
- Post from your sources: Cron → Content Sources → Cluster topics (full article on) → AI prompt (plan) → AI prompt (write) → Quality check (faithfulness only, with repair) → Publish Own Post. Put Languages between the two AI prompts for one post per language. No rule-based gate here on purpose, but the faithfulness check stays, because a chain like this may publish without anyone reading the draft: every claim is checked against the sources, a claim the sources do not support is rewritten out, and a post that still fails never reaches Publish (it waits in Content as a draft to review, with the reason). The post never names the outlets: the sources are its material, not its citations.
- One news item, many sources, an article and two posts: Content Sources → Cluster topics → AI prompt (the article) → Publish Own Post on LinkedIn (fit: always) → Publish Own Post on X (fit: always). The ready-made template does exactly this.
- Bridge from LinkedIn to X: Account posts (your Page) → AI prompt → Publish Own Post on X, fit to channel always.
- Follow back on X: New followers → Follow.
- Watch a topic on X: Cron → Posts on a topic → Post Type? → React to a post, or AI prompt → Comment on a post to reply.