Help and FAQ

Type your question into the message box and send it. A Grounded bot answers only from the knowledge and tools its owner gave it — it does not draw on general knowledge outside that. If a bot is gated, you may need to sign in or enter an access code first; see The bot is asking me to sign in or enter a code.

Beyond plain text, a bot may also support, if its owner turned them on:

• Attachments. Add a document, image, audio, or video as context for the conversation — see Attach a file, photo, audio, or video.
• Voice mode. Speak your question and hear the answer read back.
• Web search. The bot can pull in live search results alongside its own knowledge.
• Artifacts. Charts, SVGs, or small interactive results render in a side panel next to the chat.

None of these change what the bot is allowed to talk about — every answer still has to trace back to the bot's own knowledge or an enabled tool.

A Grounded bot is deliberately closed-world: it answers only from the knowledge its owner gave it, and refuses everything outside that scope rather than guessing. When your question falls outside its knowledge, you'll see its refusal message — often literally "I don't have that information," though an owner can write their own wording on the bot's Role page.

This is working as intended, not a bug: a refusal means the bot could not support an answer from its sources, so it said so instead of making one up. If you think the bot should know the answer, that's useful signal for its owner — owners can review refused and unsupported questions on their Analytics page to decide what to add next.

When a bot answers from its knowledge, look for the Sources line under the answer — it lists exactly what the bot drew on, so you can check the claim against the original material yourself.

Behind the scenes, every answer also passes through an automatic check: numbers, prices, percentages, dates, and quoted text in the answer are re-verified against the source text the bot actually retrieved for your question, before you ever see it. Some bot owners turn on an additional, stricter check that re-reads the whole answer for support. Either way, an answer with nothing to cite is marked as such rather than presented as grounded fact.

If a bot's owner has turned on file uploads, you'll see an attach control in the message box. You can add a document, image, audio, or video clip as context for that one conversation — the bot's owner decides which file types are allowed and how large a file can be.

Attached files are not added to the bot's permanent knowledge base; they only inform the conversation you attached them to. They are stored temporarily and automatically removed after about two hours (see the Privacy Policy).

Some bots restrict who can chat with them. If you land on a gate, one or both of these will apply:

• Sign in. Use your Grounded account (email/password, or a provider like Google) — see Consumer sign-in vs. Bot admin if you're not sure which sign-in you need.
• Enter a code or password. This is either a shared code the bot's owner gave everyone, or a credential set up just for you: a personal password, a one-time code, or an invite link emailed to you. If your invite link isn't working, see My invitation link isn't working.

If neither works, the bot's owner controls access — ask them to check your entry on their Users page.

Grounded has two separate sign-ins, for two different things:

• Sign in (/login/) is your Grounded account — email and password, or a provider like Google. Use it to chat with a bot that requires sign-in, and to reach My Bots, where every bot on your account lives.
• Bot admin (/admin/login/) opens one specific bot's console with that bot's own slug and admin password — separate from your account password. It's the classic way in, and still works for standalone bots that aren't linked to any account.

If a bot is on your account, you don't need its separate admin password at all: sign in to your account and open the bot's console from My Bots with one click.

What to do depends on which sign-in you're locked out of:

• Your Grounded account, signed in with a provider (Google, etc.). There's no separate Grounded password to lose — recover access through that provider directly, then sign in again the same way.
• Your Grounded account, signed in with email and password. Use password reset from the sign-in page.
• A bot's admin password. There's no self-service reset for it. If the bot was created under your Grounded account, skip the password entirely — sign in to your account and open the bot from My Bots. If it's a standalone bot not linked to any account, ask whoever set the original admin password.

Sign in and open My Bots. Every bot you created, or were made an owner or editor of, lives there: open its admin console with one click, see its monthly usage and cost, and create new bots.

To bring in help, invite a teammate by email as an owner (full admin access) or an editor (can edit content and settings, more restricted than an owner) — no separate admin password required, since they sign in with their own account.

Two ways in, both landing in the same place:

• Create a bot. Sign in, name the bot, and pick its web address.
• My Bots. Once signed in, start new bots from your dashboard — see Manage every bot on your account for the rest of what lives there.

Set the bot's persona, scope, and refusal message on its admin Role page — the refusal message is what it says, verbatim, when asked something outside its knowledge (see What does "I don't have that information" mean?).

The knowledge base (KB) is the bot's whole world, and it answers only from what is here. On the admin Knowledge page you can:

• Write entries directly, a title plus content, searchable the moment you save.
• Ingest a URL or pasted text. The page is fetched and cleaned, then the digest step drafts KB entries from it for you to review and publish.
• Ingest a file. PDFs and text extract their content; images, audio, and video are read or transcribed. All of them flow through the same digest and draft-entry review.

Draft entries stay invisible to the bot until you approve them, so nothing enters its world unreviewed.

Upload a CSV on the admin Data page (up to 5,000 rows by 30 columns) and the bot gains a query tool over it, which is good for product tables, price lists, and schedules. Column types are inferred automatically; give the dataset a clear name and description so the bot knows when to reach for it. Re-uploading under the same name replaces the data.

Everything beyond the KB is a toggle on the admin Tools page. Each capability is off by default, and per row you decide whether it is mandatory, optional for visitors, or on by default:

• Web search. Lets the bot augment answers with live search results.
• File uploads. Visitors can attach documents, images, audio, or video as one-conversation context (you choose the allowed types and size) — see Can I attach a file, photo, audio, or video to a chat? for the visitor-facing view.
• Image generation. The bot can create images on request.
• Voice mode. Hands-free speech in, spoken answers out.
• Artifacts. Charts, SVGs, and runnable HTML render in a side panel.
• MCP servers and skills. Connect external tools and packaged instructions (see the MCP and Skills pages).

A bot with only the KB enabled stays strictly grounded: it answers from its knowledge and nothing else.

The chat can be embedded on your own site, guarded by an origin allow-list: on the admin Access page, add each website origin (for example https://example.com) that may embed the bot. Requests from origins not on the list are refused; visiting the bot's own page directly always works. Per-bot rate limits apply either way.

Embed working from the bot's own page but not from your site? See My embedded bot won't load on my website.

By default a bot is open to anyone with the link. The admin Access page adds two independent gates — enable either or both, and a visitor passes if they satisfy any enabled method:

• Require sign-in. Visitors must sign in with a consumer account.
• Require access code. Visitors need a code or credential — either the bot-wide shared code you set here, or a credential you issue an individual person (see Manage who can access your bot).

Choosing a visibility of Invite or Private turns on a gate automatically if none is active yet, so the setting can't silently leave the bot open.

The admin Users page is where you manage individual people, separately from the bot-wide gate settings on Access. For each person you can set a personal password, a one-time access code, or send an invite link that lets them set their own password on first use.

Revoking a user on this page cuts their access immediately. If someone says their invite link isn't working, see My invitation link isn't working — it's usually already used or expired, both of which need a fresh invite from this page.

• Public. Listed in the public directory.
• Link (default). Unlisted; anyone with the link can chat.
• Invite. Only people you have invited.
• Private. Gated: sign-in or access code required.

Choosing Invite or Private automatically turns on an access gate if none is active, so the setting cannot silently leave the bot open. Set on the admin Access page — see How do I gate my bot?.

The admin Test Questions page holds your golden set. Each row is a question plus what you expect: phrases the answer must contain, or that the bot should refuse. Run all asks the live bot every question and marks each pass or fail, so you can catch regressions after editing knowledge, switching models, or changing the bot's role.

The admin Analytics page shows daily volume, token usage, and, most usefully, gaps: real questions the bot refused, could not support from its knowledge, or answered with "I don't have that information," grouped so you can see what to write next. Fixing gaps is the fastest way to make a bot better.

Bot refusing more than it should, or giving the wrong answer instead of just missing content? See My bot refuses everything, or says it doesn't know things it should.

An invite link looks like /join/?t=<token> and is good for one use. What you see tells you what's wrong:

• "This invite link has already been used." Someone (probably you, earlier) already redeemed it. If you already have an account for this bot, open its chat and sign in instead.
• "This invite link is invalid or has expired." Ask the bot's owner for a new one from the Users page — invite links don't renew themselves.
• "This invite link is missing its token." The link was copied or forwarded incompletely; ask for it again, and open it directly rather than retyping it.
• "Could not reach the server." A connection problem on your end — check it and try again.

Setting your password needs at least 6 characters. If the form still won't submit, the error above the button names the specific problem.

A grounded bot only answers what its knowledge and role support, so this is almost always one of three things:

• Thin or missing knowledge. Check the admin Knowledge and Data pages — if the bot hasn't been given the information, it correctly refuses rather than guessing (see What does "I don't have that information" mean?).
• A scope that's narrower than you expect. The admin Role page sets what the bot is allowed to discuss; a tight scope refuses on-topic-sounding questions on purpose.
• A real gap. The admin Analytics page groups real refused and unsupported questions so you can see exactly what to add next — the fastest way to make a bot better.

Run your golden questions on the Test Questions page after any change, to confirm you fixed it without breaking something else — see How do I test a bot before shipping a change?.

Embedding is guarded by an origin allow-list. On the admin Access page, add the exact origin your site is served from (for example https://example.com — scheme and host, no path). Requests from an origin not on the list are refused; visiting the bot's own page directly always works regardless of this list. Per-bot rate limits still apply either way, so a sudden spike in traffic can also look like an outage — check the admin Analytics page if the allow-list already looks correct.

Everything about plans and payment is in one place: sign in, open My Bots, and look under Plan & billing. What you are seeing tells you what to do:

• "Payment is past due." A charge didn't go through. Your allowance keeps working during the short retry window, so fix the card on file before it closes — Update your card opens the billing portal straight at that.
• Answers have stopped, or the usage bar is full. You've reached this month's token allowance. It resets at the start of the next calendar month; to carry on before then, upgrade the plan or buy a one-off pack under Token top-ups. Nothing else about a bot is affected — its knowledge, settings, and access gates stay exactly as configured. See Pricing for what each plan allows.
• You need to change your card, switch plan, cancel, or find an invoice. Manage billing opens Stripe's billing portal, which does all four. The card on file and your recent invoices are also listed, read-only, under Payment & invoices.
• "Could not load billing", "Could not start checkout", or "Could not open billing portal." The payment provider couldn't be reached just then. Nothing was charged and nothing changed — reload the page and try again.
• "Contact us to upgrade" where an upgrade button should be. Card payment isn't configured for this deployment, or that plan has no checkout set up yet. Ask the site owner to make the change for you.

Card details are never typed on this site. Every payment happens on Stripe's own hosted pages, and no card number is stored here.