Question 1

What is PitHub?

Accepted Answer

PitHub is a versioned workspace for professional prompts and AI conversations, Git-style history and forks for the work teams currently pass around as screenshots or loose documents.

Each saved prompt or conversation is called a pit. A pit keeps the full thread intact, attributes each reply to its AI tool, and can live in a private team workspace or a personal/public library.

Engineering, finance, sales, product, operations, and other business teams can reuse the process behind an answer instead of keeping only its final output.

Question 2

What exactly is a “pit”?

Accepted Answer

A pit is one reusable prompt or AI conversation. It has a title, optional context, tags, a professional category, and an ordered list of user and assistant messages.

Pits are versioned. When you append new messages to an existing pit you create a new version rather than overwriting the old one, so the history of how a prompt evolved is preserved.

Anyone viewing a public pit can read the full thread, copy any message, fork it into their own account or export it as Markdown, JSON, plain text or a ready-to-run prompt.

Question 3

How do I create my first pit?

Accepted Answer

There are two ways. From the web app, open Create, paste or type your prompt or conversation, add a title and tags, then choose a personal visibility or a team workspace.

From your AI agent, connect the PitHub MCP server once and say “push this conversation to PitHub” or name a team. Your agent saves the thread directly, no copy-paste and no screenshots.

Most people start on the web to see how a pit looks, then switch to the MCP flow because it is far faster once it is wired up.

Question 4

Is PitHub free to use?

Accepted Answer

Yes. Creating an account, saving pits, browsing the public library, forking, creating teams, and connecting the MCP server are free.

You need an account to create or join a team, save private work, fork, or bookmark. Reading public pits works without signing in.

Question 5

Can I publish only part of a conversation?

Accepted Answer

Yes, and you usually should, you rarely want the entire transcript. Tell your agent what you want and it maps your words onto a selector that is applied deterministically.

Say “pit the last reply”, “pit the last 2 messages”, “pit that exchange”, “pit messages 3 and 7”, “pit 2-6” or “pit just my prompts”. The confirmation tells you exactly which messages were included before anything is published.

On the web you can select the messages you want to keep directly in the editor.

Question 6

Why does selecting a reply also include my prompt?

Accepted Answer

By default, selecting an assistant reply pulls in the question above it. A published answer with no prompt is not much use to anyone, the prompt is the context that makes the reply worth reading.

If you genuinely want only the reply, say “just the reply” to turn that behaviour off.

Question 7

Can I edit a pit after publishing it?

Accepted Answer

Yes. You can change the title, description, tags and visibility at any time from the pit page or with the update_pit tool.

You can also append new messages to continue the conversation. Appending creates a new version and keeps the earlier ones in the pit's history, so nothing you published is silently lost.

Question 8

How do tags and topics work?

Accepted Answer

Tags describe what a pit is about, a language, a tool, a technique, a domain. They are how your pit gets found in search and grouped into topics on Explore.

Add a few specific tags rather than many vague ones. Topics on the home page and Explore are built from tags, so accurate tags put your pit in front of the right readers.

Question 9

What is the PitHub MCP server?

Accepted Answer

It is a small server that lets any MCP client, Claude, Cursor, Codex, Gemini and others, publish conversations to PitHub without leaving your agent.

Once connected, you can say things like “push this to PitHub”, “add this to that pit” or “search PitHub for prompts about X”, and the agent calls the matching tool. The server is a thin client over the same public REST API the web app uses.

Question 10

How do I install the MCP server?

Accepted Answer

There are two ways in, with the same tools. Remote points your client straight at pithub.app/mcp with nothing to install. npm runs the pithub-mcp package locally over stdio.

The docs page has a tailored config snippet for each supported agent, pick your agent and paste it where that agent keeps its MCP servers.

Question 11

Which tools does the MCP server expose?

Accepted Answer

The full set covers the pit lifecycle: create_pit, append_to_pit, update_pit, publish_pit, get_pit, list_my_pits, search_pits, fork_pit and delete_pit. list_teams and list_team_pits provide team workspace access.

There are also preview_pit_selection to see which messages would be published before publishing, collaboration tools like add_collaborator and list_collaborators, and list_pit_versions to read a pit's full push history.

For the files a pit carries (skill files, prompts, playbooks), add_pit_files attaches them, list_pit_files lists them, and get_pit_file reads their text back, either one named file or all of them at once.

Question 12

Is there a REST API I can call directly?

Accepted Answer

Yes. The MCP server is a thin client over a public REST API, so you can build against it directly. Authenticate with an Authorization: Bearer <token> header.

Endpoints cover pits (create, read, update, delete, fork, like, bookmark, export), the feed, Explore, users and the current account. Create a token in Settings and it is scoped to your account and revocable at any time.

Question 13

How do I sign in?

Accepted Answer

You can sign in with email, with Google or with GitHub. All three land you in the same account, pick whichever you prefer.

You only need to be signed in to publish, fork, like, bookmark or manage tokens. Reading public pits does not require an account.

Question 14

How does connecting my agent stay secure?

Accepted Answer

With the npm server, asking your agent to connect opens an approval page in your browser. You sign in there, click Authorize, and the client receives a scoped API token. Your password never touches the client.

With the remote server you create a token yourself in Settings and put it in the Authorization header. Either way the agent only ever holds a scoped, revocable token, never your credentials.

Question 15

Where do I manage my API tokens?

Accepted Answer

In Settings. You can create tokens there and revoke any of them at any time.

Revoking a token immediately stops every client using it. If you think a token has leaked, revoke it and create a new one, nothing else needs to change.

Question 16

Can I change my username?

Accepted Answer

Your username is part of your public profile URL at /u/<username>, so it appears on every pit you have published.

You can update your profile details in Settings. Because links to your old username may exist elsewhere, change it deliberately rather than often.

Question 17

What visibility options does a pit have?

Accepted Answer

Personal pits can be public, unlisted, or private. Public pits appear in the library and search; unlisted pits are link-only; private pits are limited to their owner and explicitly added collaborators.

Team pits use a stricter rule: they are always private to current team members, and their visibility cannot be changed.

Question 18

Is a private pit really private?

Accepted Answer

Yes. Private and unlisted pits are excluded from public listings, search and the sitemap, and access is enforced at the database level with row-level security, not just hidden in the interface.

The anonymous client the site uses for public pages physically cannot read a private row, so a private pit cannot leak into a feed or a crawler even by mistake.

Question 19

How do I keep secrets out of a published pit?

Accepted Answer

Publish only the messages that matter. Using a selector to include just the relevant exchange is the simplest way to leave setup, keys and unrelated chatter out of a pit.

Review the selection confirmation before publishing, and remember you can start a pit as private, clean it up, and only then switch it to public.

Question 20

Can I delete a pit completely?

Accepted Answer

Yes. You can delete any pit you own from its page or with the delete_pit tool, which removes it and its versions.

Forks are independent copies, so deleting your original does not delete forks other people made earlier. If a pit is sensitive, set it private or delete it promptly rather than relying on obscurity.

Question 21

How do I find good pits?

Accepted Answer

Explore is the main entry point. It groups public pits by topic and surfaces what is active, so you can browse by the subject you care about rather than scrolling one long list.

Search lets you find pits by keyword and tag across everything public.

Question 22

What is the difference between Home and Explore?

Accepted Answer

Home is the latest professional library: a quick view of newly shared public pits. Explore adds full-text search, professional categories, tags, and sorting.

Private team work never appears in either place. Open the Team tab to browse a workspace you belong to.

Question 23

How do interests shape what I see?

Accepted Answer

During onboarding, and later in Settings, you can choose your areas of work. They provide shortcuts to the categories and tags relevant to you.

They do not affect team access: membership alone determines which team pits you can read.

Question 24

What do bookmarks and likes do?

Accepted Answer

A bookmark is a private shortcut to a pit so you can find it again from the Bookmarks page.

Forking is the more important work signal in PitHub: it creates an editable copy while preserving lineage to the source.

Question 25

What does forking a pit do?

Accepted Answer

Forking copies someone else's public pit into your own account as an independent pit you can edit, extend and republish.

It is how you build on a conversation, take a prompt that nearly worked, fork it, refine it and share your improved version, with the lineage back to the original preserved.

Question 26

How do teams collaborate on pits?

Accepted Answer

Create a team workspace and add collaborators by PitHub username or account email. Every member can read, create, edit, append, fork, and comment on pits in that team.

Only the team owner can add or remove members or delete the workspace. Removing a collaborator immediately removes access to every team pit.

Question 27

How does version history work?

Accepted Answer

Every push to a pit, yours or a collaborator's, creates a new version rather than overwriting the last one. The pit always shows the latest, but the earlier versions are kept.

You can read the full history with list_pit_versions, which is how a prompt's evolution stays visible instead of being flattened into a single final state.

Question 28

Is the original author credited when I fork?

Accepted Answer

Yes. A fork records where it came from, so the lineage from your version back to the original stays intact.

That attribution is the point, it lets a good prompt spread while still crediting the person who started it.

Question 29

My agent can't connect to PitHub, what should I check?

Accepted Answer

First confirm the MCP config is in the right place for your agent; each agent keeps its servers in a different spot, and the docs snippet is tailored per agent.

For the remote server, check that your Authorization: Bearer token is present and has not been revoked. For npm, re-run the connect step so the browser approval page can issue a fresh token. Ask your agent to run check_connection to confirm who it is connected as.

Question 30

The wrong messages ended up in my pit. How do I fix it?

Accepted Answer

Use preview_pit_selection before publishing to see exactly which messages a selector will include, that catches most mistakes up front.

If a pit is already published, update it or append the messages you meant to include. If it is beyond repair, delete it and publish again with the correct selection.

Question 31

Why isn't my pit showing up in search or Explore?

Accepted Answer

Only public pits appear in search, Explore and the sitemap. Check the pit's visibility, if it is unlisted or private it is reachable by link (or not at all) but will not be listed.

Public listings and the sitemap are recomputed periodically rather than instantly, so a brand-new public pit can take a little while to appear. Adding accurate tags also helps it surface.

Question 32

I'm still stuck, where can I get help?

Accepted Answer

Start with the MCP and API docs, which cover installation, connection and every tool in detail.

If something looks broken rather than merely confusing, note what you asked your agent, what it did and any error it returned, that context makes it far quicker to sort out.

Question 33

How is context engineering different from prompt engineering?

Accepted Answer

Prompt engineering focuses on instructions; context engineering manages the complete information environment used by the model.

Question 34

What is the board?s role in AI governance?

Accepted Answer

The board oversees strategy, material risk, accountability, capital allocation, resilience, and whether management?s controls and evidence are credible.

Question 35

What should an employee AI policy include?

Accepted Answer

It should cover approved tools, data handling, human review, intellectual property, disclosure, prohibited uses, retention, incidents, training, and enforcement.

Question 36

What is the best AI tool for a small business?

Accepted Answer

The best tool fits a specific bottleneck, existing workflow, budget, data rules, and employee capability. Start with the problem, not the vendor.

Question 37

What should an AI support agent automate first?

Accepted Answer

Start with frequent, stable, low-risk questions and guided workflows that have clear escalation paths.

Question 38

What is the difference between SEO and GEO?

Accepted Answer

SEO targets search discovery broadly; GEO emphasizes being understood and cited in generated answers. Their technical and content foundations substantially overlap.

Question 39

What is AI search optimization?

Accepted Answer

It is the practice of making useful content easy for AI answer systems to discover, understand, verify, and cite.

Question 40

Should AI SaaS charge by token?

Accepted Answer

Usually not as the primary customer-facing metric unless the buyer is technical and token use maps clearly to value.

Question 41

What makes AI UX different from traditional UX?

Accepted Answer

AI outputs vary and can be wrong, so the interface must communicate uncertainty, support correction, and manage autonomy and trust.

Question 42

Are embeddings personal or sensitive data?

Accepted Answer

They can encode sensitive source information and should inherit appropriate classification, access, retention, and deletion controls.

Question 43

Can prompt injection be completely prevented?

Accepted Answer

No single control eliminates it. The practical goal is defense in depth that prevents untrusted text from causing material harm.

Question 44

What is an LLM eval?

Accepted Answer

An eval is a repeatable test that measures a model or application against explicit criteria using representative inputs and known expectations.

Question 45

Is AI-generated code less secure?

Accepted Answer

It can be secure or insecure, but its speed and apparent fluency make disciplined security review especially important.

Question 46

Should coding agents work directly on the main branch?

Accepted Answer

No. Use an isolated branch or worktree so changes can be inspected, tested, and discarded safely.

Question 47

What is the formula for generative AI ROI?

Accepted Answer

A useful starting point is net attributable benefit divided by total AI program cost, with benefits adjusted for adoption, quality, and risk.

Question 48

Is MCP secure by default?

Accepted Answer

MCP is a protocol, not a complete security system. Production security depends on authentication, authorization, tool design, deployment, and client behavior.

Question 49

Is RAG better than fine-tuning?

Accepted Answer

Neither is universally better. RAG supplies current knowledge; fine-tuning changes stable behavior. They can be used together.

Question 50

What is a good success rate for an AI agent?

Accepted Answer

There is no universal threshold. Required performance depends on consequence, human review, reversibility, and the quality of the alternative process.

Question 51

Who should own AI agent governance?

Accepted Answer

Business owners should own outcomes, technology owners should own operation, and security, privacy, legal, and risk teams should define and test controls.

Question 52

What is the best first AI agent use case?

Accepted Answer

A high-volume, rules-bounded workflow with measurable outcomes, reliable data, and reversible actions is usually the strongest first use case.

Question 53

Can prompt injection attack an MCP tool?

Accepted Answer

Yes. Untrusted content can influence a model to call tools incorrectly, so permissions and server-side validation must not depend on the prompt alone.

Question 54

Should AI agents replace entire jobs?

Accepted Answer

Start with tasks and workflows, not entire jobs. Most roles combine automatable work with judgment, accountability, and relationship-heavy responsibilities.

Question 55

Do low-risk AI agents need human approval?

Accepted Answer

Not necessarily for every action. Approval should be proportional to consequence, uncertainty, sensitivity, and reversibility.

Question 56

Can an LLM judge another AI agent?

Accepted Answer

Yes, but only after calibration against expert labels and with monitoring for systematic grader bias.

Question 57

Does a large context window replace RAG?

Accepted Answer

Not usually. Long context helps with bounded document sets, while retrieval controls relevance, permissions, freshness, and cost across larger corpora.

Question 58

How do you value time saved by AI?

Accepted Answer

Count time only when it is verified and converted into additional output, avoided hiring, faster revenue, better service, or another observable outcome.

Question 59

What makes a good coding-agent task?

Accepted Answer

A bounded outcome, clear constraints, relevant context, acceptance criteria, and known verification commands.

Question 60

Should reviewers read every line of AI code?

Accepted Answer

Review depth should match risk, but reviewers must understand every behavior and side effect they approve.

Question 61

How many evaluation examples are enough?

Accepted Answer

Enough to cover meaningful task and risk slices with stable estimates. Start focused, then expand from production failures and new use cases.

Question 62

Does more context improve AI answers?

Accepted Answer

Only when the additional context is relevant, trustworthy, well ordered, and within an effective attention budget.

Question 63

What is indirect prompt injection?

Accepted Answer

It is an attack embedded in content the model retrieves or observes, such as a webpage, document, email, or tool result.

Question 64

Should AI prompts be retained?

Accepted Answer

Retain them only for a defined purpose and period, with controls appropriate to the data they may contain.