postecho：基于 Claude Code 的自托管 X 与 LinkedIn 写作助手

Find a topic and draft posts from your writing examples.
A self-hosted writing assistant for X and LinkedIn. You review and schedule each post yourself.

Who it's for: people who post on X and LinkedIn and want to run their own writing assistant. Each install has one owner. Writing requires Claude Code logged in with your Claude plan. Jev optionally ranks results and gives feedback on writing style. The preview above uses fictional sample data.

What it does

Find. Search sources including Hacker News, GitHub and arXiv. Some sources need credentials. Jev ranks results when enabled; search also works without it.

Turn a video into posts. Get twelve X drafts from a YouTube transcript. If unavailable, postecho can use the description and chapters, or you can paste a transcript. Check facts and add attribution before publishing.

Compose. Choose an idea for three drafts based on your writing examples and style guide. Pick one and ask for changes, such as a shorter version or a LinkedIn post.

Schedule. Add finished posts to the queue in Compose. Schedule lines them up at your next posting times and opens X and LinkedIn for each in turn, already filled in. You set the time there and record it in postecho. If you change that time on the platform, update postecho too.

What makes it different

Your writing examples guide the drafts. Proposed style-guide updates wait for your approval.
Optional style feedback. Jev scores AI-like writing patterns. Humanize tries up to three rewrites and keeps the best-scored result. The score is a model's assessment.
Visible progress. Searches and writing jobs show their current step and elapsed time.
Your Claude Code login. The local agent uses your Claude plan, subject to its usage limits.
Manual publishing. You review each draft and schedule it in the platform's own composer.
Local hosting. Run the app and database on your computer, or deploy the web app separately.

Built by

Simone Lovera, CPO at Forte AI, designed and built postecho with Claude Code. Read the design case study.

Run it

You need Git, macOS or Linux, and Node 22.9 or newer. To generate or revise text, install Claude Code and log in with your Claude plan.

git clone https://github.com/simojam93/postecho && cd postecho
npm run setup   # installs dependencies and prepares the local database
npm run demo    # optional: adds fictional sample data
npm run dev     # starts the web app and local agent

Open the address npm run dev prints, http://localhost:3000 unless another program already uses that port, and log in with the password chosen or generated during setup. With sample data loaded, open Compose to inspect the example drafts. Generating new text requires Claude Code.

For a first live task, add writing examples in Settings, search a topic in Find Ideas, and use a result to generate three drafts. Choose one and edit it in Compose.

Add optional service keys in Settings. For online hosting, see the deploy guide.

What it costs

postecho's code is free. Connected services have their own costs and limits.

	What for	Cost
Claude	Generating and revising text	Your Claude plan and its usage limits
Jev by TypeSafe	Optional ranking and style feedback	TypeSafe's pricing
X's API	Optional X search	X's API pricing
Vercel, Neon	Optional online hosting	Provider plans and limits; Vercel Hobby is for personal, non-commercial use

How it works

flowchart LR
  you([You]) --> web["Web app<br/>Next.js on Vercel"]
  web --> db[("Postgres<br/>Neon, or PGlite locally")]
  web -- "puts jobs" --> queue[["Job queue"]]
  agent["Agent on your computer"] -- "takes jobs" --> queue
  agent --> claude["Claude Code<br/>claude -p"]
  web --> jev["Jev, through jev-judge<br/>rank · spam · human score"]
  web --> sources["Open sources<br/>Hacker News · GitHub · Bluesky · arXiv · …"]

The web app queues jobs. An agent on your computer runs Claude Code and returns the results. The page wakes the agent on your computer when it creates a job, so the agent sends nothing while PostEcho is closed. Keep that agent running for writing jobs, including when the web app is hosted online. Text used for writing goes to Claude; enabling Jev sends text for evaluation to TypeSafe.

Jev integration uses jev-judge, an MIT library included in web/vendor/. See Contributing to add another writer.

Folder	What's in it
`web/`	Web interface, API and database access.
`agent/`	Local worker that runs Claude Code.
`scripts/`	Setup, development and sample-media scripts.
`docs/`	Deployment, case study, specs and plans.

Stack: TypeScript, Next.js 16, React 19, Tailwind CSS 4, Drizzle, Postgres/PGlite, Vitest, Claude Code and Jev.

Tests

npm test

Runs the web, agent and setup-script tests with local data and test doubles. No service keys are needed.

Contributing

Help with a bug, a source adapter, setup or another AI writer. Start with CONTRIBUTING.md, follow the code of conduct, and report security issues through SECURITY.md.

License

MIT. See LICENSE.