Frequently Asked Questions

DoThat is AI infrastructure for businesses that want to adopt AI, not build it — the cloud model applied to AI. Just as the cloud let any company run software without a data centre, DoThat lets you deploy your own AI assistants ("Goblins") — grounded in your knowledge and given your voice — for your teams and your customers, in days, with no ML stack or engineers. The enterprise-grade machinery (isolation, access control, per-deployment cost caps) runs underneath so you don't have to.

No. The platform is built for the non-technical business — a function or ops lead can stand up an AI assistant without writing code. You give your Goblin its identity, write its instructions in plain English, and upload your documents; DoThat handles the underlying infrastructure, knowledge retrieval, and AI routing.

Yes — those are the two modes we serve. For your teams, give Support, Sales, Ops, Legal, and HR each their own assistant, grounded in their documents and run from one place. For your customers, put a branded AI assistant in front of your users — isolated from everything else and on a budget you set. The same platform serves both; you choose the outcome, not a topology.

In days, not quarters. There is no infrastructure project to scope and no ML team to hire. Configure and test a Goblin in preview first — refine its instructions, model, and Datasets privately — then publish it and add a deployment for the audience you choose.

You can upload standard formats — PDF, DOCX, TXT, and MD — into your workspace's Data, and group them into Datasets. The platform indexes them automatically, and a Goblin answers from the Datasets and files you link to it — your material, not generic web content.

It answers from your own knowledge and in your own voice, rather than giving generic answers. You ground each Goblin in your private Datasets and shape its behaviour directly — instructions, personality, and model choice — and you can set your Organization logo.

Yes. Invite team members into your Workspace and they collaborate based on the roles you assign. Owners and Admins manage plans, members, and settings across the organization; everyone else joins as a Member and is granted Editor or Viewer access per Workspace. Editors build and deploy Goblins; Viewers have read-only access.

Sharing a Goblin grants chat-only access and is separate from management roles. You can share via a public link, a secret link, an internal or external group, or named people. Whoever you share with reaches only the chat screen — never the Goblin's setup, prompt, or Datasets.

Yes. Add external people — customers or contractors — to a magic-link group, and they verify their email once to chat with your Goblin. They are strict chat-only participants with no platform account and no view of your configuration.

Isolation is the core of the platform. Your knowledge is grounded privately and never pooled into generic answers, data never crosses between Organizations, and the management controls and chat access run as two separate engines — so the people you share a Goblin with reach only the chat, never your prompts, configuration, or Datasets, and one audience can never reach another's.

Your chat and knowledge-retrieval requests are processed on Google Cloud Vertex AI, which — per Google — does not use them to train Google's models. (This covers the chat and retrieval path; image generation runs on a separate Google service.) We also do not grant any AI provider the right to train models on your uploads, prompts, configurations, or conversations.

DoThat is a pre-release beta — it is not yet a live, generally-available service. It is operated by DoThat AI Ltd, registered in England and Wales (company number 17351077). While it is in beta, platform data (your account, Goblins, Datasets, and conversations) may be deleted at any time, without notice, as part of beta operations, and the service is provided as-is with no warranties or guarantees of any kind — on every plan, paid or free. Please don't use it to store anything you can't afford to lose, and keep your own copies of anything important. See our Terms and Privacy Policy for the full detail.

Goblin Gold is our transactionally exact metered unit of AI work. Every chat turn deducts Goblin Gold from a plan instance, and each instance carries its own monthly allowance, an absolute lifetime cap, and a top-up pool. You can give different teams or audiences separate budgets, so one high-volume user can't run up everyone's costs. We meter and cap spend per team and audience; the platform never bills your end customers.

We cover the core enterprise controls: strict tenant isolation built for GDPR and privacy obligations, encryption at rest and in transit, database-backed role-based access control, Model Armor screening of model inputs, and a standalone platform-admin plane with a full audit log of administrative actions. We are built to SOC 2 and ISO 27001 standards, though we do not yet hold those formal attestations.

Hosted single sign-on (including Google and passkeys) is available today via our identity provider, WorkOS. Enterprise SSO (SAML/OIDC) and SCIM directory sync can be configured for you on request during onboarding.

Your Organization is the single payer — one plan, one bill, in your chosen currency (GBP, USD, or EUR). A Plan Template in our catalog defines default limits (Goblins, seats, storage, Goblin Gold); when your Organization buys one, it mints a plan instance with those limits snapshotted onto it, which you can assign to a Workspace to give it its own budget.

Yes. Consultancies, IT partners and product teams build on DoThat to deliver AI for the businesses they serve — the same way cloud integrators build on a cloud platform. Stand up isolated AI per client, each with its own cost control, all under one account. You keep the client relationship and set your own price.

That is one of the most common ways in. The logic, the prompts and the steps transfer; what gets replaced is the scaffolding around them — access control, audit, budgets, credentials, checks and review gates. Describe what the workflow does and the Workflow Assistant drafts it for you to correct, test in a sandbox and release.

Yes. Everything available in the interface is available over the API and MCP, so you can provision, publish and run agents from your own systems, in your own pages, on your own domain.

The models, the cloud, the storage and retrieval, the security tooling and the people who watch it, penetration tests and the audit programme, transactional email, patching and monitoring, and support. There is no separate model-provider account, cloud bill, vector database licence or sysadmin to hire alongside it.

The shared platform runs in a US region and serves UK and EU customers under standard contractual clauses and the UK addendum. Where in-region hosting is a hard requirement, we deploy a dedicated instance in your region — its own data store, storage, compute, keys and logs — scoped and dated in the contract.

No. Your content is not used to train models, ours or anyone else's, and we commit to that contractually rather than only on a web page.

Not yet — both are in progress. The control set those frameworks ask for is implemented and evidenced today, and an independent penetration test is scheduled. What we do not hold yet is the report and the certificate. Ask us for the evidence pack; it is usually enough for a vendor-security review to proceed with a dated condition.

It stays yours throughout and is isolated to your organisation. Deletion is two-stage — withdrawn immediately, then permanently purged after a fixed retention window — with an explicit action to erase deliberately rather than waiting.