Questions we get asked.
Straight answers, including the unflattering ones. Where the honest answer is “no”, it says no — and says what we do instead.
What it costs, and why we won’t guess.
What does it cost?
There is no corporate price list, because there is nothing honest to put on one yet. What a deployment costs depends on how many people you put on it and what you connect, so we quote it properly on a call instead of guessing on a web page. The only prices we publish anywhere are for different products on a different host — the $99/month individual membership on buildwithsumit.com, and the usage credit packs for our job-application tool. Neither is this portal.
See the pricing page ›Why won’t you publish a price?
Because the meter does not exist. There is no seat count, no seat limit, no per-seat billing and no corporate invoicing built into the product today — the payment integration we do have covers only the individual membership and one unrelated credit pack. Publishing a number would commit you to a deal shape we have not built. We would rather tell you that than print a figure.
Is there a trial?
Not a self-serve one. There is no sign-up flow that provisions a company: your organisation, your domains and your portal host are created by hand before anyone can sign in. So a pilot is something we set up with you and agree the shape of on the call, not a button on this page.
How do our people get accounts?
They sign themselves in. Once your domain is registered against your portal, anyone with an email at that domain gets a 6-digit code and is enrolled on first sign-in — no password, no invite list to maintain. The domain check is exact equality, not a suffix test, so a lookalike domain does not match. Today that is the only way in: there is no SSO and no directory sync.
How sign-in works ›Who supports us?
The people who wrote it — Sumit and the Globussoft team. There is no support tier, no ticket portal and no published response time, and we are not going to invent one to fill this answer. What you get is direct access to the team that ships the code, on terms we agree on the call.
How long does a rollout take?
It depends on what you connect, and we will not quote you a number we have never measured. Connecting an account is a Google sign-in; reading a large Drive and the mail window into a searchable memory is a background job that runs for as long as the volume takes, and the summarised memory built on top of it is rebuilt on a 20-minute cycle after that. We have not timed a full corporate onboarding end to end, so we have no honest “live in N days” to give you.
Short answers. The whole answer is on one page.
Each answer here is the compressed version of something written out in full, with the same numbers, on the security page.
Where is it hosted?
One Ubuntu server in Oracle Cloud’s Mumbai region (ap-mumbai-1). The site, the databases, the extracted document store, the generated notes and the on-box backups are all on that one machine. There is no replica and no failover.
Where your data sits ›Is it encrypted at rest?
Partly, and this is where most vendors would round up. In transit it is TLS. Google refresh tokens are encrypted in the database. Database backups are encrypted before they leave the box. Everything else — extracted document text, email bodies, chat history, the generated notes — sits unencrypted on the server’s disk: no disk-encryption layer, no encrypted database tablespaces.
The precise statement ›Who else sees our data?
Inside the product, only the person it belongs to: the identity comes from a signed session cookie, every tool is called with that identity, every query filters on it in SQL, and a read by file id is ownership-checked, not just filtered. The exception is the shared company vault, which is labelled to the model as shared precisely because everyone in the deployment can see it. Outside the product, your text reaches Anthropic when the assistant answers; if that path is unavailable a turn can fall through to DeepSeek, which is configured and live today; and voice audio reaches ElevenLabs. There is no per-tenant control over any of that and no sub-processor list. And Sumit has root on the server.
Who can see what ›Can we get an EU region?
No. One machine, one region, Mumbai. There is no residency choice, no second region and no way to pin your data to a geography. If EU residency is a hard requirement, the honest route today is to run it yourself — the code is public and AGPL-3.0 — rather than to buy this hosted portal.
Hosting and residency ›Do you have SOC 2?
No. No SOC 2, no ISO 27001, no HIPAA, no data-processing agreement, no sub-processor list and no penetration-test report. We have never been audited by anyone. If procurement needs a certificate, there is none to send you — and no badge on this site pretending otherwise.
What we do not have ›Do you support SSO?
No. No SAML, no SCIM, no OIDC, no identity-provider integration, no directory sync and no automated provisioning. Sign-in is a 6-digit code to your work email, on a domain you have registered with us. That is the whole authentication story.
Sign-in and access ›Can we see an audit log?
No. There is no admin-visible log of which employee asked what, which documents the assistant read, or who ran which agent. The server records sync runs and flags suspicious-looking inputs, but nothing renders either of those to you, and neither is an access trail. Do not buy this expecting auditability.
The gaps, listed ›How do we remove someone who leaves?
By hand, by us, on the server — and this one deserves the blunt version, because it is the sharpest gap in the product for a corporate buyer. The seat record has statuses for suspended and removed, and no code path writes either. The admin console can grant or revoke an agent and set someone’s team and role; it cannot remove a seat. Until that is built, offboarding is a request to us and a manual database update.
Offboarding, in full ›Can we get our data out?
Not through a button. A person can disconnect a Google account and delete a connected source, but nothing today purges the extracted files, the generated notes, the chat history or the summary built from them. Export and deletion are manual, done by us on the server, on request — so treat them as something we agree with you, not a feature you can point at.
Retention and deletion ›What it does. What it doesn’t.
Can it send email on our behalf?
It drafts, you approve. The one send path is deliberately narrow: the assistant can send a draft it already wrote and you already confirmed in the conversation, by passing back that draft’s own id — it cannot compose and send free text. The email agents that work a shared support desk write into that mailbox’s drafts for a human to review and send — today they run on our own desks, and putting them on yours is something we wire up with you. One agent does mutate a mailbox, and we would rather name it than bury it: it moves a wrongly spam-filtered customer back into the inbox.
What the agents do ›What can it actually read?
On this corporate portal: each employee’s own Google Drive, Gmail and Analytics, connected by them, through your portal’s own Google client. From Drive — Docs, Sheets, Slides and the text formats, plus uploaded PDF, DOCX, XLSX and PPTX. From Gmail — the last 90 days by default, spam and trash excluded. Images, video, audio and archives are skipped outright, and there is no OCR, so a scanned PDF honestly reads as empty. Telegram, WhatsApp, Teams, Obsidian and CRM ingest exist in the wider product but are not wired on the corporate host.
Connector by connector ›Does it understand meaning, or match words?
Retrieval matches words. There are no embeddings and no vector search anywhere in it: filenames are matched with a SQL LIKE and token ranking, and content search is a literal substring scan over the most recently extracted files. A synonym misses. What carries the understanding is the summarised memory of your business, pasted verbatim into the model’s prompt on every turn — that is where the meaning lives, and it is rebuilt on a schedule rather than at the moment you ask. It is a deliberate choice, and calling it semantic search would be a lie.
How retrieval actually works ›Is there a mobile or Slack app?
No. No Slack app, no Teams app, no mobile app, no desktop app. It is a page in a browser at your own portal address, and that is the entire surface.
Does voice work for our staff?
Not on the corporate portal. Voice is built into the individual member page — the speech runs on ElevenLabs while the brain stays ours, so a spoken turn is wired to the same memory, the same tools and the same saved preferences as a typed one — but corporate chat is text-only today. Being straight about that feature: it is deployed and configured, and it has carried no traffic in the last month, so treat it as built rather than proven. If someone shows you a voice orb on a corporate page, they are showing you the wrong page.
Text and voice ›Can we run it ourselves?
Yes. The stack is public at github.com/Build-With-Sumit/globus under AGPL-3.0, with a Dockerfile, a compose file, install notes, the database schema and tests. Yours to run. Yours to audit. What we will not claim: there is no supported on-prem edition, no enterprise self-hosted SKU, and we have not timed the install for you — so if you go this way, budget your own engineering time for it.
The limits have their own pages.
Every page on this site ends with what that part cannot do. They are not footnotes and they are not buried — they are the part you have to plan around, so read them before you decide, not after.
Read these before you buy
- What it cannot read — skipped file types, no OCR, the 90-day mail window, and which connectors exist everywhere except here. Connectors
- What retrieval really is — word matching, not meaning; a memory rebuilt on a cycle, not a live one. Product
- What the agents will and will not do — drafts, review queues, and an agent console with nothing in it yet. Agents
- Where your data sits, and what is not encrypted — one machine, one region, no certification, no audit log, no self-serve offboarding. Security
- Why there is no price — no seats, no meter, no corporate billing built yet. Pricing
You will also notice there are no customer logos, testimonials or case studies on this site. Globussoft builds Globus on itself, and the desks you can see running on it are our own — we would rather show you that than dress it up as someone else’s proof.
Ask us the one we didn’t answer.
Bring the awkward question — the procurement one, the residency one, the “what happens when someone leaves” one. You will get the same answer on a call that you got on this page, with the working shown.