Skip to content
Pitchbox

It researches and drafts. You send. Nothing leaves without you.

A self-hosted outreach agent for Reddit, Hacker News, Mastodon and LinkedIn.

The checks behind every draft

  • It reads what you've already written before it claims to know your voice. Under three pieces of writing, it says so and stops there, rather than inventing a style.
  • A checker in the code strips tells like the em dash before a draft reaches you. No model is asked nicely to skip them - the code just does it.
  • If a tell survives that pass anyway, the draft still shows it to you instead of hiding it.
  • Nothing sends on its own. The worst a rough draft can do is sit in your inbox until you delete it.

How it works

  1. 1 It researches the thread and drafts a reply in your voice.
  2. 2 You read it in the Inbox and approve, edit or reject it.
  3. 3 You send it yourself, and Pitchbox records that you did.
The Pitchbox inbox showing three drafts waiting for review, one opened with its reasoning and an Approve button.
Drafts wait for you in the Inbox. Nothing sends until you approve.
The Pitchbox inbox filtered to Sent, showing messages the operator sent themselves with a sent date.
Once you send it, Pitchbox records it. It never sends for you.

The in-page companion

A Chrome extension reads the LinkedIn post you already have open and drafts a comment in the same panel style as the Inbox.

It reads the page you opened. It never clicks anything. You press LinkedIn's own button.

The Pitchbox panel open on a LinkedIn post, showing a drafted comment with Insert, Drier, Warmer and Shorter controls.
The panel that reads the post you have open, and nothing else.

What it will not do

  • It never sends a message. Every draft waits in your inbox until you approve it.
  • It never crawls LinkedIn. The in-page companion reads the post you already opened, nothing else.
  • It respects a blocklist and a dedup window, so the same person is never contacted twice by mistake.
  • It shows you the cost of every run before you approve what it drafted.

Two ways to run it

Self-host it

Run it on your own infrastructure, under the AGPL license, with your own API key.

Or use the cloud edition

The models are ours and every run carries a spending cap, so there's nothing to provision.