Not another AI text box. A content production pipeline.
A five-stage AI pipeline: analyze, plan, write, adapt, format. Most "AI for social" is a single prompt wearing a marketing page — you paste a topic, it hands back a caption. PulseCraft runs a production line instead. It reads your source, decides what to say, writes it natively for each network, aligns it to your brand, and formats it to every platform's rules — before a human ever looks.
A five-stage pipeline, not a single prompt — with AI usage included as credits.
One piece of source content becomes up to ten platform-ready posts.
SYS.001
The five stages
The engine treats every source item the way a content team would: not as one prompt, but as five jobs, each with its own context and its own output.
Stage 1 — Analyze
The engine reads the source — an article, a search result, an uploaded document, an image — and extracts what matters: the claim, the angle, the facts, the entities. Uploaded media (PDFs, images, office files) is transcribed and visually described first, so the writer works from meaning, not file bytes. It then matches the item against your subject → topic taxonomy.
INPUT: raw source → OUTPUT: structured brief + topic match
Stage 2 — Plan
Before any prose, the engine decides the shape: which content types to produce, the angle for each, and how one source should differ across networks. A single article can become a LinkedIn carousel, an X thread, and a Medium long-read — each planned to its own intent, not copy-pasted.
INPUT: brief → OUTPUT: per-platform content plan
Stage 3 — Write
Now it drafts — platform-native from the first word. LinkedIn gets a measured professional register; X gets compression to 280 characters; Reddit gets a human, non-promotional voice; Hacker News gets title conventions that don't get flagged. Each draft is written for where it lives.
INPUT: content plan → OUTPUT: native drafts per platform
Stage 4 — Adapt
Every draft passes through your three-layer brand style — visual identity, language tone, and mood — so the output sounds like you, not like a model. Set the style once; the engine applies it to every post on every platform.
INPUT: drafts → OUTPUT: brand-aligned drafts
Stage 5 — Format
Finally, mechanics: character limits, hashtag conventions, media specifications, aspect ratios, and the platform's accepted file types. The post that reaches the queue is already valid for its destination.
INPUT: brand-aligned drafts → OUTPUT: publish-ready posts
SYS.002
One source, ten posts
Feed the engine a single article and it fans out into genuinely different posts — not the same caption reformatted, but native copy for each network.
The same source. Ten native posts. Zero copy-paste.
10 platforms: LinkedIn, Facebook, Instagram, X (Twitter), TikTok, YouTube, Pinterest, Reddit, Medium, and Hacker News.
SYS.003
Content type routing
12 content types — from text posts and carousels to long-form articles, Reels, and Shorts. The plan stage routes each source to the content types that fit it and the platforms you've enabled:
| Content type | Typical destination |
|---|---|
| Text post | LinkedIn, Facebook, X, Reddit |
| Image + caption | Instagram, Facebook, LinkedIn |
| Carousel | LinkedIn (PDF), Instagram |
| Thread | X (Twitter) |
| Long-form article | Medium, LinkedIn articles |
| Reels / short video | Instagram, TikTok, YouTube Shorts |
| Video | Facebook, YouTube, TikTok |
| Image / video pin | |
| Link post | |
| Tell / Ask / Show HN | Hacker News |
SYS.004
AI usage included as credits
AI usage included — predictable credit pricing, no metering surprises.
Generation runs on the built-in engine and draws from your plan's credit balance — there's no provider account to wire up and no metered model bill that swings month to month. On Enterprise, the entire platform self-hosts, so content stays within your own infrastructure.
Read how credits work in transparent AI pricing: credits, not marked-up keys, or see the Integrations page for the full architecture.
SYS.005
Built for power users
- Aggregated mode. Combine several source items into one synthesized post — a weekly roundup, a themed digest — instead of one post per item.
- Custom prompts. Override the engine's instructions per workflow when you need a specific angle, structure, or constraint the defaults don't cover.
Generic AI assist writes a caption. An engine runs a pipeline.
A chatbot answers a prompt. A pipeline analyzes, plans, writes, adapts, and formats — and does it across ten networks on a schedule. See the difference in PulseCraft vs Buffer or the full alternatives landscape.
Questions, answered
Which AI models does the engine use?
Each pipeline runs on leading models, with usage included as credits. If you'd rather run your own, you can bring your own keys for Gemini, GPT, or Claude and point the engine at them — your account, your model choice.
Can I edit the prompts the engine uses?
Yes. Each workflow accepts custom prompt instructions, so you can steer the angle, structure, or constraints when the defaults aren't enough.
Does the output sound robotic?
The adapt stage runs every draft through your three-layer brand style — visual identity, language tone, and mood — and the write stage drafts natively per network rather than reformatting one generic caption. The result reads like your team, not a template.
What data does the AI see?
Only the sources you connect and the brand styles you define. On a self-hosted Enterprise deployment, that content stays within your own infrastructure.
Forty minutes from now, this could be running.
Free month · 10 posts · No credit card
Related: How it works · Brand Intelligence · Integrations · Multi-step AI pipelines vs one-shot prompts · AI credit pricing · Pricing