About PitHub
Your best prompts should outlive the chat window.
PitHub is a versioned library for the prompts, AI conversations, workbooks, and skill files behind your team’s work. Capture what worked once, group it, improve it in versions, and reuse it instead of rebuilding the same context from scratch every time.
- Versioned, never overwritten
- Private by default in teams
- Saves straight from your agent
- Built to be reused
On this page
The short version
What PitHub is
If you only read one paragraph, read this one.
PitHub is a versioned library for prompts and AI conversations. Each saved unit is called a pit: the full conversation, the files it produced, the context needed to find it again, and the history of every change made to it. Pits can be private to you, private to a team, or published to a public professional library.
What a pit is not
Not a screenshot
A screenshot records that something worked. A pit records how, in a form you can run again, edit, and hand to someone else.
Not a snippet in a doc
A pasted prompt loses the replies that proved it works, the files it produced, and every revision since. A pit keeps all three together.
Not a chat history
Scrollback is a log, ordered by when you happened to type. A library is ordered by what the work is about, and it is searchable months later.
Why it exists
The problem with prompts as scrollback
Every team using AI seriously runs into the same four things within a few months. None of them are model problems; they are all storage problems.
The good prompt is lost by Friday
It took eleven turns to get right and it lives in one person's scrollback, in one tool, under a title nobody chose. Next month it is faster to rewrite it badly than to find it.
The same context gets re-explained
Every fresh thread starts by re-teaching the model the same background. That is time spent, tokens spent, and a different answer every time.
Improvements do not accumulate
Someone sharpens the prompt in their own window. Nobody else sees it, so the next person starts from the old version, or from nothing.
There is no shared standard
Two people ask for the same deliverable and get two different shapes of answer, because there is no agreed prompt to point at.
Code solved this decades ago and nobody argues about it any more: you keep the source, you version the changes, and you let other people fork it. Prompts are now doing real work, and they deserve the same treatment.
The core loop
The pit lifecycle
A pit is not filed away and forgotten. It moves through five stages, and the last two feed each other for as long as the prompt is useful.
- Stage 1 · Capture
Keep the thing that worked, whole
A pit is created from the conversation itself, not a summary of it. Every message keeps its role and the model that wrote it, so the reasoning is still legible later. Save it from the editor, paste a transcript, or say “pit this” to a connected agent.
- Stage 2 · Organize
Put it where it will be found
A title, a description, a category, and up to ten tags. Personal pits sit on your profile; shared ones live in a team workspace. This is the step that decides whether the pit is findable in three months.
- Stage 3 · Improve
Change it without destroying it
Append a better exchange, edit the working version, or fork it to try a different approach. Each push is a new version with a note about what changed and who changed it. The previous version stays readable.
- Stage 4 · Reuse
Start from your best, not from blank
The next person opens the current version, changes the inputs, and runs it. They inherit the context, the constraints, and the phrasing that already earned a good answer, and they never re-derive any of it.
- Stage 5 · Measure
Let usage tell you what to keep
Forks, comments, bookmarks, and views show which prompts a team actually reaches for. The ones nobody touches can be retired; the ones in constant use are worth another pass.
The unit of work
What you actually keep in a pit
“Prompt library” undersells it. The things worth saving from AI work are rarely just one paragraph of instructions.
Prompts
The instruction that finally produced the right answer, kept with the replies that proved it, so the next person can see what “right” looked like.
Workbooks
The documents a conversation produced or depends on: briefs, board updates, specs, rubrics, checklists. They travel with the pit rather than living in a separate drive folder.
Skill files
Reusable instruction files your agent can load. Keeping the skill next to the conversation that shaped it means both are versioned together, and neither drifts alone.
Playbooks
Multi-step procedures, review passes, and interview scripts, kept as an ordered thread so the sequence matters and nothing gets skipped.
Shared, not scattered
Grouping the work
A library only works if the right people can reach it and the wrong people cannot. Grouping in PitHub happens on three levels.
Team workspaces
A private home for the prompts a team shares. Members can create, edit, fork, attach, and comment; a removed member loses access immediately.
Categories and tags
One category places a pit in the right corner of the library; up to ten tags carry the finer detail, and both are searchable.
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.
A pit that belongs to a team is structurally private: it cannot be switched to public or unlisted, because the constraint lives in the database rather than in a settings screen.
Compounding
Improving together, without losing the past
This is the part that turns a folder of prompts into an asset. A prompt that four people have sharpened is worth more than four private copies of the first draft.
Every push is a version
Appending an exchange, editing the working copy, or attaching a file snapshots the pit as the next version, with a note about what changed and who did it.
Fork to diverge
Copy a pit, with its conversation and files, into your own account and take it somewhere else. The link back to the original is preserved on both sides.
Discuss in place
Threaded comments sit on the pit itself, so the argument about why v3 beat v2 is stored next to v3, not in a chat thread nobody can find.
Read the lineage
Version history shows each push in order, so a newcomer can see how an approach evolved rather than inheriting only its final state.
The payoff
Reuse, and what it saves
A saved prompt earns its keep the second time it is used. Everything before that is filing; everything after is compounding.
Fewer wasted turns
Most of the cost of a repeated task is the turns spent re-establishing context and converging on a phrasing that works. Starting from a pit skips them.
A predictable answer
The same prompt produces the same shape of deliverable across a team, so reviews stop being about format and start being about substance.
Onboarding that is real
Handing someone the prompts your team actually uses teaches more, faster, than any document describing how the team works.
We deliberately do not publish a “saves you N% of tokens” number. The saving is real but it depends entirely on your prompts, your models, and how often a pit is reused. What PitHub guarantees is the mechanism: the context you already paid to write is still there next time.
Where it plugs in
It meets you in the tool you are already in
The fastest way to use PitHub is never to open PitHub. Connect the MCP server once and saving becomes one sentence to your agent.
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” or “pit just my prompts”. Selectors are applied on the server and named back to you, so sensitive setup stays out.
Take it anywhere
Continue a pit in Claude, ChatGPT, or Gemini in one click, or export it as Markdown, JSON, plain text, or a bare prompt.
The audience
Who PitHub is for
Anyone whose good AI work is currently trapped in a chat window they will never scroll back to.
Engineering teams
Keep the agent prompts, review passes, and skill files that work, and stop three people from writing the same one privately.
Business and operations
Standardise recurring deliverables like updates, briefs, and summaries on one prompt everyone can point at and improve.
Researchers and consultants
Carry a method between engagements with its lineage intact, and show how an approach evolved instead of only its final state.
Solo builders
One searchable library instead of four tools' worth of scrollback, and a public profile for the work you want found.
How we build it
What we hold to
Three commitments that decide most of the product arguments before they start.
Private means private
Visibility is enforced in the database, not hidden in the interface. A pit inside a team is always private and cannot be switched to public. That is a rule in the schema, so it holds for the web app, the API, and every MCP client alike.
How privacy is enforced →History over cleanliness
Nothing is silently overwritten. Every push snapshots the conversation and the file list, because the version you regret is often the one you need to read next week.
How versioning works →No lock-in
Any pit exports as Markdown, JSON, plain text, or a bare prompt, and continues in Claude, ChatGPT, or Gemini in one click. A library you cannot leave is not a library.
Read the API and MCP docs →Straight answers
Questions people ask first
Short answers to the things worth knowing before you sign up. The full FAQ goes considerably deeper.
What is PitHub?
PitHub is a versioned library for the prompts and AI conversations behind real work. Each saved unit is called a pit: it holds the full conversation, any files it produced, the context needed to find it again, and the history of every change made to it.
What is a pit?
A pit is one unit of useful AI work, kept whole: an ordered thread of messages with each reply attributed to the model that wrote it, plus attached text files, a title, description, category and tags, and a version history. It can be public, unlisted, or private to you or your team.
How is PitHub different from just saving prompts in a document?
A document keeps the text of a prompt. PitHub keeps the conversation that proved it works, the files it produced, and every revision since, and it lets a teammate fork it without disturbing your copy. It is also searchable by category and tag, and reachable from your AI agent directly.
Does reusing prompts save tokens?
Reusing a saved prompt removes the turns normally spent re-explaining context and converging on a phrasing that works, and those turns are where most tokens go on a repeated task. How much you save depends on your own prompts and how often they are reused, so PitHub does not claim a fixed figure.
Can I keep my prompts private?
Yes. Every pit is public, unlisted, or private, and pits inside a team workspace are always private to current members. Visibility is enforced by database rules rather than by the interface, so it holds across the web app, the API, and any connected AI agent.
Can I save a pit straight from Claude, ChatGPT, or another agent?
Yes. PitHub runs an MCP server, so a connected agent can create pits, append to them, attach files, and search your library from inside the tool you are already working in. Claude, ChatGPT, Cursor, Codex, Gemini, VS Code, and Windsurf are supported.
What can I attach to a pit?
Text artifacts: Markdown, txt, JSON, JSONL, YAML, CSV, TSV, XML, and TOML, up to 10 MB each and 50 files per pit. Attaching a file records a new version, and files inherit the pit's visibility and travel with it when it is forked.
Does PitHub cost anything to read?
No. Public pits are free to read without an account, and every public pit is also available as Markdown, JSON, or plain text from its export endpoint.
Keep the prompt behind the work
Make your first pit in a minute, or connect your agent so saving is one sentence away.