How it works
Git-style history for prompts and AI conversations
PitHub gives technical and business teams a shared place to keep the prompts behind their work, improve them together, and retain the history instead of passing screenshots around. This page walks through the whole product, start to finish.
- Step 1
Capture the useful work
Turn a prompt or AI conversation into a pit, the full thread, with every reply attributed to the tool that wrote it. Save from the web editor or straight from your agent over MCP.
- Step 2
Build in a team workspace
Create a private workspace, add collaborators, and keep its pits available only to current team members. Every member can create, edit, fork, and comment without making the work public.
- Step 3
Version and fork improvements
Append new results, edit the working version, or fork an approach without losing its source. PitHub keeps the lineage and version history so the team can see how a prompt evolved.
On this page
The core idea
What a pit is
A pit is one unit of useful AI work, kept whole. Not a screenshot and not a copied snippet, the conversation, the files it produced, the context needed to find it again, and the history of how it changed.
The conversation
An ordered list of messages, each tagged as user, assistant, system, or tool, and each able to record which model wrote it. Markdown and fenced code blocks render as written.
Files
Text artifacts the conversation produced or depends on, skill files, prompts, playbooks, workbooks. They travel with the pit and are versioned alongside it.
Context
Title, description, the tool it came from, the model, a category for Explore, and up to ten tags. This is what makes a pit findable months later.
History
Every push snapshots the conversation and the file list as a new version, so nothing is overwritten and you can read how the work evolved.
Step 1
Create a pit
Two routes in. Use the editor when you are working in the browser, and your agent when the conversation is already happening somewhere else.
Write or paste it
The editor builds the conversation message by message, marking each as yours or the AI's. Give it a title, a description, a category, and tags so it can be found later.
Import a transcript
Paste a raw transcript and PitHub parses it into messages, filling in the title, model, and tags where it can. The same parsers back the MCP server, so both routes behave identically.
A pit needs a title and at least one message. Everything else can be filled in later, editing a pit is a normal part of using it, not a correction.
Step 2
Attach files
Conversations often produce something reusable: a skill file, a prompt template, a playbook, a workbook. Those belong with the pit rather than in a separate drive folder.
What you can attach
Text formats, md, markdown, txt, text, json, jsonl, yaml, yml, csv, tsv, xml, and toml. Up to 10 MB each and 50 files per pit.
How they get there
Attach them in the editor when creating a pit, from the Files panel on the pit page afterwards, or have your agent push them over MCP.
They are versioned
Attaching a file records a new version, so the file list is part of the pit's history rather than a mutable side channel.
They follow the pit's privacy
Files are stored privately and served through short-lived signed links, so a private or team pit's attachments stay as private as the pit itself.
Re-attaching a file under a name that already exists replaces its contents. Files belong to the pit that carries them and travel with it when it is forked.
Step 3
Version and fork
Prompts get better by being changed. PitHub is built so that changing one never destroys what it used to be.
Every push is a version
Creating, appending, editing, or attaching a file snapshots the conversation and file list as the next version, with a note about what changed and who did it.
Append as you learn
When a prompt produces a better result, append the new exchange rather than starting over. The pit stays one continuous record.
Fork to diverge
Forking copies a pit, conversation and files, into your account so you can take it somewhere else. The link back to the original is preserved on both sides.
Read the lineage
Version history shows each push in order, so you can see how an approach evolved instead of only its final state.
Step 4
Work as a team
Most valuable prompts are not meant to be public. Teams are the shared, private home for them.
Team workspaces
Create a team, invite collaborators, and its pits are visible only to current members. Removing someone removes their access immediately.
Everyone can build
Any member can create, edit, append, fork, attach files, and comment on the team's pits. Versions record who made each push.
Per-pit collaborators
Outside a team, the owner of a single pit can invite specific people to push to it, by username or by email.
Move work in
An existing personal pit can be shared into a team, which makes it private to that team and hands access to its members.
Step 5
Control who sees it
Visibility is set per pit and enforced in the database, not merely hidden in the interface.
Public
Discoverable in Explore, search, and feeds. Anyone can read and fork it.
Unlisted
Readable by anyone with the link, but never surfaced in Explore or search.
Private
Only you, your invited collaborators, or your team can read it.
Team pits are structurally private
A pit that belongs to a team is always private and cannot be switched to public or unlisted. That is a rule in the database, not a setting in the interface, so it holds for the web app, the API, and every MCP client alike.
Privacy details →Step 6
Save from your agent
The fastest way to use PitHub is never to open PitHub. Connect the MCP server once and publishing becomes one sentence in the tool you are already in.
Connect once
Add the hosted server at pithub.app/mcp, or install the npm package for stdio clients. Claude, ChatGPT, Cursor, Codex, Gemini, VS Code, and Windsurf are all supported.
Then just ask
“Pit this conversation.” “Add the playbook to that pit.” “Find my onboarding pit and fork it.” The agent calls the same API the web app uses.
Publish exactly what you mean
Say “pit the last reply”, “pit 2-6”, or “pit just my prompts”. Selectors are applied deterministically on the server and named back to you in the confirmation, so unrelated or sensitive setup stays out, and you can preview the selection before anything is published.
Step 7
Discover and reuse
A prompt library is only worth keeping if you can find things in it again, yours and everyone else's.
Explore
Browse public pits by trending, new, popular, or recommended, and narrow by category or tag.
Your feed
Latest shows everything new, Following tracks the people you follow, and For You draws on the categories and tags you have chosen to track.
Fork it
Copy someone's pit into your own account, conversation and files, and change it freely. The link back to the original is kept.
Take it elsewhere
Continue a pit in Claude, ChatGPT, or Gemini in one click, copy it as Markdown, or export it as md, json, txt, or a bare prompt.
Step 8
Signals and standing
Public pits accumulate signals that tell you which ones actually earned their keep.
Likes and views
Lightweight signals of what a pit is worth. Views are counted once per viewer, not per refresh.
Bookmarks
Save any pit you can read to your own list, including ones you did not write.
Comments
Discuss a pit in threaded replies, one level deep. Comments can be liked too.
Pitting power
A standing earned from the engagement your pits receive, likes, forks, comments, and views, weighted. It starts at 100.
Follow people whose work is useful and their pits appear in your Following feed. None of this applies to private or team pits, which stay out of every public surface.
Ready to start?
Make your first pit in a minute, or wire up your agent so publishing is one sentence away.