Write a content template once, then turn any topic into everything it needs: blog posts, social media posts and captions, first comments, newsletters, video scripts, product reviews, launch campaigns, hashtags and SEO. Written in your brand voice by Claude, ChatGPT, Gemini or a local Ollama model. Self-hosted, team-ready, MIT licensed.
Quick start · Templates · Features · Screenshots · How it works · FAQ · Docs
Built by dotmirror
dotcontent is a free, open-source, self-hosted AI content generator for any kind of content. You describe a piece of content once, as a template: what it is for, the rules it follows, and the sections it is made of. Then every topic you add, whether a keyword, a product, an idea, a link or a video, becomes the whole set in one click.
One template might write a blog post with its outline, meta description, FAQ and promotion posts. Another writes a week of Facebook statuses, Instagram captions, LinkedIn posts and the first comment under each. Another writes a product review in three formats, or a full launch campaign. Every piece comes out in the same brand voice, and every section can read the ones written before it, so nothing contradicts anything else.
It is made for bloggers, social media managers, agencies, creators, newsletter writers, affiliate marketers and small marketing teams who are tired of pasting the same prompt into a chat window every day.
Seven example templates ship with the app, each a different kind of content. Run them as they are, or copy one and make it yours.
| Template | Use it for | What you get |
|---|---|---|
| SEO Blog Post | Blogs, SEO, content marketing | Keyword and intent research, outline, titles and meta description, the full article, FAQ, social posts to promote it |
| Social Media Post Pack | Daily posting, social media management | Facebook statuses, Instagram caption, LinkedIn post, X posts and thread, first comments, hashtags |
| Honest Product Review | Affiliate sites, review channels, e-commerce | Verdict with pros and cons, video script, written review, caption with first comment, YouTube copy |
| Newsletter Email | Newsletters, email marketing | Issue plan, 10 subject lines with preview text, the email, a subscriber post |
| Repurpose a Video | Podcasts, YouTube, webinars | Takeaways and quotes, clip ideas, blog article, LinkedIn, Facebook and X posts, Instagram carousel |
| Product Launch Campaign | Brands, agencies, launches | Launch brief, teaser posts, launch-day posts, launch and reminder emails, video script, ad copy |
| Website Shorts | Reels, YouTube Shorts, TikTok | English and Bangla scripts, on-screen captions, hooks, social captions, hashtags, YouTube SEO |
Read the template guide → It covers what every template writes, how to run one, how to write your own, and more ideas: YouTube long-form, podcast show notes, product descriptions, cold email, case studies and more.
Build a template in three steps, without code. Drag in sections (Article,
Social post, Script, Caption, Email, First comment, Hook, Title, Description,
Hashtags, CTA, Research, Plan, or your own), write each one's instruction,
and set the order. Use placeholders like {{topic}} and {{series}}.
Export a template as a .template.json file and import it on another
console.
Sections run in dependency order. Research is written before the article that cites it, the outline before the draft, and the posts before the first comments that sit under them. Sections that do not depend on each other are queued together, so several machines can write them at once. A failed section costs only that section, and you can rewrite it on its own.
Sections are written by the claude (Claude Code), codex (ChatGPT) or
gemini CLI already signed in on your computer. They are billed to your
existing subscription, not per token. An Anthropic API key or a fully local
Ollama model works too. Each workspace picks its own
model.
Each workspace has its own brand voice, languages, goal and model, and every section is written under that voice. Run the same template for two clients and get two different voices. It writes in any language the model writes well; the shipped Website Shorts template writes English and Bangla.
Paste a video link (YouTube, Reels, TikTok, a podcast clip) and the worker downloads it with yt-dlp, cuts stills with ffmpeg and transcribes it with whisper.cpp. Make a topic a web page's link and the server reads the page. Either way, the content is written from the source, not from guesses.
The console runs on one server. The writing happens on teammates' own laptops, which claim jobs from a queue. Close the tab and the run carries on; if a laptop goes to sleep, another machine picks the job up. Invite teammates by link, and manage machines and tokens in Settings. An optional Windows desktop app bundles the console and a worker.
Your topics, prompts and content stay in your own Postgres. API keys are encrypted with AES-256-GCM. Deploy with Docker Compose (Traefik and Let's Encrypt included) or Dokploy.
 Templates: seven examples, one for each kind of content |
 A template: the Social Media Post Pack, section by section |
 Builder: drag in Article, Social post, Email, First comment and more |
 Content: topics grouped by template, one click to run |
 Home: the current workspace and what is waiting on you |
 Integrations: Claude, ChatGPT, Gemini and Ollama, probed live |
You need:
| Node.js 24 | Check with node --version. The worker runs TypeScript directly, which needs Node 22.18 or newer. |
| Docker | For Postgres. Docker Desktop on Windows and macOS. |
| A model | At least one of: Claude Code (claude), the Codex CLI (codex) or the Gemini CLI (gemini), signed in. Or an Anthropic API key, or Ollama. |
| bash | start.sh is a bash script. On Windows, use Git Bash. |
git clone https://github.com/shimentosh/dotcontent.git
cd dotcontent
cp .env.example .env
./start.shstart.sh installs the dependencies, starts Postgres in Docker (port 5437),
the API on http://localhost:4000 and the console on http://localhost:3333.
-
Open http://localhost:3333 and sign up. The first account becomes the owner, and after that sign-up closes. Invite the rest of the team from Settings.
-
Add this computer as a machine. Go to Settings → Machines, add one and copy the token (it is shown once). Put it in
.env:DOTCONTENT_WORKER_TOKEN=wrk_…
Stop
start.sh(Ctrl+C) and run it again. It starts a worker as well. -
Check your model. Integrations shows each CLI and tool the machine can reach. Test sends a real one-line prompt.
-
Try a template. Every sample workspace has topics waiting under a different template. Pick a workspace, open Content, and press Run on a topic. Then set your own brand voice in Workspaces and add your own topics.
-
Optional, for video:
npm run worker:setupdownloads pinned, checksummed yt-dlp, ffmpeg and whisper.cpp into.data/tools. It does not touch your PATH or need admin rights.
browser ──► console (Next.js, :3333)
│
└──────► API (NestJS, :4000) ──► Postgres
▲
│ HTTPS: claim a job, post the result
│
worker(s) on your team's machines
└─ runs claude / codex / gemini / ollama, yt-dlp, ffmpeg, whisper
Every section's prompt is built in the same order: the workspace's brand voice, then the template's purpose, then its rules, then the topic, any source page or video, and the sections it depends on.
- Press Run. The API queues one job for each section whose inputs are ready.
- A worker claims it. It runs the model CLI on its own machine, with that machine's login, and posts the finished text back.
- The next sections are queued as soon as what they depend on is done.
- Read, edit, copy, download. The reader groups sections into the deliverables the template declares: Blog post, Promotion, Emails.
The screens say Template where the code says pack, and Machine where it
says worker.
- Bloggers and SEO writers: keyword research to published article, with the FAQ and promotion posts done too.
- Social media managers: a post for every platform from one idea, with the first comment and hashtags, every day.
- Agencies: one workspace per client, each with its own voice, and templates you reuse across all of them.
- Creators and podcasters: turn one video into clips, an article and a week of posts.
- Affiliate and e-commerce sites: honest reviews in video, blog and social formats from one set of facts.
- Marketing teams: a launch campaign, from teasers to the reminder email, that tells one consistent story.
docker-compose.prod.yml runs Postgres, the API, the console and Traefik with
Let's Encrypt. docker-compose.dokploy.yml is the same app for Dokploy. Start
from .env.production.example. A server has no signed-in CLI, so either set
ANTHROPIC_API_KEY or leave the writing to your teammates' machines.
docs/DEPLOYING.md has every step.
What kinds of content can dotcontent write?
Any text content: blog posts, social media posts and statuses, captions, first comments, hashtags, newsletters and emails, video and podcast scripts, product descriptions and reviews, ad copy, YouTube titles and descriptions, and anything else you can describe in a template. It does not generate images or audio.
Is dotcontent free?
Yes. dotcontent is open source under the MIT License: free to use, change and self-host, including commercially. You pay only for the model you choose, and with a CLI you already subscribe to there is no extra cost.
Do I need an OpenAI or Anthropic API key?
No. By default sections are written by the claude, codex or gemini CLI
signed in on a teammate's computer. An API key is optional. It is useful on a
server, or as a fallback when no machine is awake, and that fallback is off
until you turn it on.
Can I run it fully offline with a local model?
Yes. Point a workspace at Ollama and the writing, including whisper.cpp transcription, runs on your own hardware.
Which languages does it support?
Any language the model writes well. Templates are plain instructions, and the example templates write in whatever language the workspace's brand voice asks for. The Website Shorts template writes English and Bangla.
Can I write my own templates?
Yes. That is the point. Start from scratch in the builder, or open one of the examples and change it. Templates export and import as JSON files, so a team can share them. See the template guide.
Does it post to Facebook, Instagram or my blog for me?
Not yet. dotcontent writes the content and you publish it with the tools you already use.
| Templates | Every example template, how to run one, and how to write your own |
| How it works | The longer tour: the data model, sign-in, keys and video ingest |
| Architecture | What lives where, the routes, and the data flows |
| Worker | The job queue, leases, machine tokens and the desktop app |
| Developing | Running it, checking it, and the traps |
| Deploying | Putting it on a server for a team (Docker Compose or Dokploy) |
| Decisions | Calls already made, and why |
app/ one page per route (Next.js App Router)
components/ the screens; components/ui is the UI kit every screen is built from
lib/ client state, theme; lib/packs holds the shipped templates
lib/server/ everything that touches the database
api/ the NestJS API, sharing lib/server with the root
worker/ the sidecar on a teammate's machine. It has no database.
src-tauri/ the optional Windows desktop app: the console plus a worker
tests/ vitest, pure functions only
docs/ the design, written down
Issues and pull requests are welcome, and so are new templates: export yours from the builder and share it in an issue. Read CONTRIBUTING.md, then run the checks before you open a pull request:
npm run check # typecheck, lint, tests, and the API's typecheckPlease follow the Code of Conduct. Report security problems privately, as described in SECURITY.md.
dotcontent used to be called Content OS. Every environment variable is now
DOTCONTENT_*. The old CONTENTOS_* names are still read when the new one is
not set, so existing .env files and deployments keep working. The Postgres
role, database and volume keep the name contentos, because renaming stored
data needs a migration, not a find-and-replace. Upgrading also adds the six
new example templates to your library; delete any you do not want and they
stay deleted.