Confidential today. More possible tomorrow.
Forge reads, checks and drafts your documents and runs your repeatable processes, under rules your organisation sets, with a record of what it did.

Built for confidence
Five questions to ask before you trust AI with real work, and how Forge answers each of them.
We call this governed delegation: AI does the work, inside the limits you set, and leaves a record of what it did.
Explore Our SecurityAnswers cite the document they came from, and totals are worked out by code over the whole file, not estimated by the model.
In projects, personal details are masked before a question reaches the model. Your data is stored in the US and never used for training.
Access is granted project by project, and administrators set privacy rules no member can weaken. Approval steps are coming.
Every access, AI call and program run is recorded, without storing what anyone wrote or asked.
Choose Anthropic or OpenAI for your organisation, and use your own key so model spend sits on your account.
Turn documents into opportunity
Automate document-heavy workflows like contract reviews, invoice processing, vendor onboarding and more, safely and accurately, at scale.
Confidential Agreement
PII Detected
and redacted
Processing with Forge...
Document processed securely
Ready for your review
Forge, by GradTensor, lets an organisation delegate work to AI with confidence. It reads, checks and drafts documents and runs repeatable processes, under rules the organisation sets, with a record of what it did.
The idea behind it →Their business tiers are good assistants for each person, with administrator controls and no-training terms. Forge puts the rules with the organisation: access granted per project, masking before any model is called, a choice of model provider, and workflows that handle every document the same way. It uses Anthropic's and OpenAI's models underneath.
ChatGPT, Claude or Copilot for business →Yes. That is what it is for. Access is granted per project, personal details are masked before a prompt leaves, uploaded originals are deleted once their text is extracted, and nothing you send is used to train a model.
Is it safe to upload them to ChatGPT? →Yes. An organisation chooses Anthropic or OpenAI, and can use its own API key so model spend sits on its own account.
Yes, for email addresses, phone numbers, card numbers, Aadhaar and PAN numbers, and written US social security numbers. Masking is on by default in projects, and an administrator can make it a rule no member can switch off. Names and addresses are not detected.
Workflows extract and check documents automatically, and a person moves each record to its next stage. Approval steps with named approvers are being built, so that routine cases can run on their own and exceptions go to the right person.
Deciding what may run on its own →No, and it is not meant to. Forge sits above your systems of record and handles the document work between them. Results can be exported, and connections that act in those systems directly are being built.
Every access and every AI call goes on an append-only record, with the model used and whether masking applied, but never the content. Each move of a record through a workflow is recorded too, and usage and cost are shown per person.