Enterprise Security, AI security on your own key
Enterprise Security protects everything your company puts into UniversalContext. Your data is encrypted with your organization's own key and kept in your own storage. AI traffic goes through a gateway you deploy, so nothing stays in a model provider's logs, and deleting the key makes every document unreadable.
Your organization's own key
Your data is encrypted with a key no other customer shares.
Access for people and agents
Each answer holds only what its asker, person or agent, is cleared to see.
Zero data retention
AI traffic runs through a gateway you deploy, never into a model provider's logs.
Crypto-shredding
Delete the key and every document, chunk, and entity becomes unreadable, after a recovery window.
Can you prove where your company's data went after the AI read it?
Only if the architecture was built to let you, and "trust us" is not an architecture.
Before phase 2 begins, Harbor Point's general counsel sends Acme a security questionnaire. Whose key encrypts our drawings, who can open our files, does any AI provider keep a copy, and what happens to our data if you stop? A file pasted into someone's personal chat account has no answer to any of them.
Where does security show up in an ordinary day's questions?
Try it: pick a question
They will accept a cap equal to the fees, drawn only from the matters you are cleared for.
Drafted from the account's own files, with nothing left in the model provider's logs.
Three accesses, each by someone cleared for it, with the time of each from the audit trail.
What happens to a file between the day it arrives and the day it is destroyed?
- Step 01
Encrypted before it touches disk
The Harbor Point services agreement arrives through a connector or an upload. Before it touches disk, it is encrypted with Acme Engineering's own key, and each chunk cut from it for search is encrypted at the application layer as well.
- Step 02
Kept apart from every other organization
The file is stored in Acme Engineering's own storage, and what the platform learns from it, the client, the project, the obligations, goes into Acme Engineering's own intelligence database. No key, no storage, and no database is shared with another customer.
- Step 03
Asked about, and kept out of the model's logs
Maya Okafor asks Nirah what the agreement says about site access. The question goes to the model through a gateway Acme Engineering deploys, under zero data retention, so the answer comes back and nothing stays in the model provider's logs.
- Step 04
Destroyed by deleting one key
When the data has to go, the key is deleted, not the records. A seven-day recovery window comes first, in case the deletion was a mistake. After it, every document, chunk, and entity encrypted with that key is cryptographically unreadable.
What stands between your files and the wrong reader?
One question, answered for whoever asks
Access is applied at the moment of the question, not after it. Ask what was decided about phase 2 staffing as the project lead, and the answer draws on the staffing plan and the kickoff. Ask as the managing partner, and it can draw on the partner compensation memo too. An agent gets only what its role allows, like anyone else.
Layer on layer, each one its own
Role-based access, row-level security, document classification levels, application-layer encryption of every chunk, your own storage and intelligence database, and your own key under all of it. Defense in depth means no single control is the only thing standing in the way.
Zero data retention, whichever model answers
Every AI call, whether Nirah answers a question or an automation runs on Monday morning, flows through a self-hosted gateway your organization deploys. Your queries never touch a third party's logging, and the rule holds with any model provider you use.
A record of who saw what
Every access is recorded in a complete audit trail. When an auditor, or a client's general counsel, asks who opened the Harbor Point site plans and when, the answer is a list, not a guess. Agents appear in it beside people.
How is Enterprise Security different from AI chat on a business plan, suite copilots, and building your own AI stack?
Compare Enterprise Security with
- Enterprise Security
- The security under everything in UniversalContext: your own key, your own storage, your own gateway for AI traffic, and access for people and agents.
- General AI chat on a business plan
- A public AI assistant on a business or enterprise plan, used across your company, under the provider's own security controls.
Encrypted with a key that is yours alone
- Enterprise Security
- Full:Your organization's own key, shared with no other customer
- General AI chat on a business plan
- Partial:Encrypted by the provider; a key you manage on some top tiers
Kept out of model providers' logs
- Enterprise Security
- Full:Every AI call through a gateway you deploy, under zero data retention, with any provider
- General AI chat on a business plan
- Partial:Not used for training on business plans; kept for a period under the provider's policy
Each person and agent sees only what they are cleared for
- Enterprise Security
- Full:Role-based access, row-level security, and classification levels, for people and agents
- General AI chat on a business plan
- Partial:Workspace roles and each connector's permissions; agents vary by product
Provably unreadable when the data has to go
- Enterprise Security
- Full:Delete the key and every document, chunk, and entity is unreadable, after a 7-day window
- General AI chat on a business plan
- Partial:Deleted when you ask, and under retention settings
A record of who saw what
- Enterprise Security
- Full:A complete audit trail of every access
- General AI chat on a business plan
- Partial:Admin and compliance logs, on enterprise tiers
Ready without a security build of your own
- Enterprise Security
- Full:Built into the platform; the AI gateway is the one piece you deploy
- General AI chat on a business plan
- Full:Switched on with the plan, under the provider's controls
- Enterprise Security
- The security under everything in UniversalContext: your own key, your own storage, your own gateway for AI traffic, and access for people and agents.
- Suite copilot
- AI built into one vendor's office suite, working across that suite's mail, files, chat, and meetings under the controls the suite already has.
Encrypted with a key that is yours alone
- Enterprise Security
- Full:Your organization's own key, shared with no other customer
- Suite copilot
- Partial:The suite's encryption; a key you hold on the tiers that offer it
Kept out of model providers' logs
- Enterprise Security
- Full:Every AI call through a gateway you deploy, under zero data retention, with any provider
- Suite copilot
- Full:Processed inside the suite's own service boundary, under its commitments
Each person and agent sees only what they are cleared for
- Enterprise Security
- Full:Role-based access, row-level security, and classification levels, for people and agents
- Suite copilot
- Full:The suite's own permissions, for people and for the copilot
Provably unreadable when the data has to go
- Enterprise Security
- Full:Delete the key and every document, chunk, and entity is unreadable, after a 7-day window
- Suite copilot
- Partial:Retention and deletion policies; key revocation where you hold the key
A record of who saw what
- Enterprise Security
- Full:A complete audit trail of every access
- Suite copilot
- Full:The suite's audit log, copilot activity included
Ready without a security build of your own
- Enterprise Security
- Full:Built into the platform; the AI gateway is the one piece you deploy
- Suite copilot
- Full:Inherits the controls your suite already runs
- Enterprise Security
- The security under everything in UniversalContext: your own key, your own storage, your own gateway for AI traffic, and access for people and agents.
- Your own AI stack in your cloud
- Models, a search index, and the code that joins them, built and run by your own team in your own cloud account.
Encrypted with a key that is yours alone
- Enterprise Security
- Full:Your organization's own key, shared with no other customer
- Your own AI stack in your cloud
- Full:Your own keys, in your own cloud account
Kept out of model providers' logs
- Enterprise Security
- Full:Every AI call through a gateway you deploy, under zero data retention, with any provider
- Your own AI stack in your cloud
- Full:Yours to arrange: self-hosted models or zero-retention terms per provider
Each person and agent sees only what they are cleared for
- Enterprise Security
- Full:Role-based access, row-level security, and classification levels, for people and agents
- Your own AI stack in your cloud
- Partial:Possible, once your team writes it into every search and every agent
Provably unreadable when the data has to go
- Enterprise Security
- Full:Delete the key and every document, chunk, and entity is unreadable, after a 7-day window
- Your own AI stack in your cloud
- Full:Yours to design: delete your own keys and your own stores
A record of who saw what
- Enterprise Security
- Full:A complete audit trail of every access
- Your own AI stack in your cloud
- Partial:As complete as the logging your team builds
Ready without a security build of your own
- Enterprise Security
- Full:Built into the platform; the AI gateway is the one piece you deploy
- Your own AI stack in your cloud
- None:Every control designed, built, and kept up by your team
| General AI chat on a business planA public AI assistant on a business or enterprise plan, used across your company, under the provider's own security controls. | Suite copilotAI built into one vendor's office suite, working across that suite's mail, files, chat, and meetings under the controls the suite already has. | Your own AI stack in your cloudModels, a search index, and the code that joins them, built and run by your own team in your own cloud account. | Enterprise SecurityThe security under everything in UniversalContext: your own key, your own storage, your own gateway for AI traffic, and access for people and agents. | |
|---|---|---|---|---|
| Encrypted with a key that is yours alone | Partial:Encrypted by the provider; a key you manage on some top tiers | Partial:The suite's encryption; a key you hold on the tiers that offer it | Full:Your own keys, in your own cloud account | Full:Your organization's own key, shared with no other customer |
| Kept out of model providers' logs | Partial:Not used for training on business plans; kept for a period under the provider's policy | Full:Processed inside the suite's own service boundary, under its commitments | Full:Yours to arrange: self-hosted models or zero-retention terms per provider | Full:Every AI call through a gateway you deploy, under zero data retention, with any provider |
| Each person and agent sees only what they are cleared for | Partial:Workspace roles and each connector's permissions; agents vary by product | Full:The suite's own permissions, for people and for the copilot | Partial:Possible, once your team writes it into every search and every agent | Full:Role-based access, row-level security, and classification levels, for people and agents |
| Provably unreadable when the data has to go | Partial:Deleted when you ask, and under retention settings | Partial:Retention and deletion policies; key revocation where you hold the key | Full:Yours to design: delete your own keys and your own stores | Full:Delete the key and every document, chunk, and entity is unreadable, after a 7-day window |
| A record of who saw what | Partial:Admin and compliance logs, on enterprise tiers | Full:The suite's audit log, copilot activity included | Partial:As complete as the logging your team builds | Full:A complete audit trail of every access |
| Ready without a security build of your own | Full:Switched on with the plan, under the provider's controls | Full:Inherits the controls your suite already runs | None:Every control designed, built, and kept up by your team | Full:Built into the platform; the AI gateway is the one piece you deploy |
Compared by category, as of November 2026. Products within a category vary; each cell says how.
What works with Enterprise Security
- 500+ Integrations

Connectors
Every file a connector brings in is encrypted with your key before it touches disk.
- Your Company's AI

Nirah
The assistant that answers inside each person's access, under zero data retention.
- Personal Brain

MyUC
A private layer for each person, visible to no colleague, manager, or administrator.
Questions about Enterprise Security
Updated
Where is our encryption key held?
Each organization's data is encrypted with its own customer master key in AWS Key Management Service. No other customer shares it. Your data is also kept in its own storage and its own intelligence database, isolated from every other organization's.
What happens when the key is deleted, and can it be undone?
The key is deleted, not the records, and every document, chunk, and entity encrypted with it becomes cryptographically unreadable. A seven-day recovery window comes first, so a mistaken deletion can be reversed; after that it is final. Crypto-shredding is compliant with the GDPR right to erasure.
Does a model provider keep what we ask?
No. All AI traffic flows through a self-hosted gateway your organization deploys, so your queries never touch third-party logging infrastructure. That holds whichever model provider answers.
Do agents follow the same access rules as people?
Yes. Role-based access, row-level security, and document classification levels are enforced at query time, whether a person asks or an agent does. A person's MyUC goes further: it is visible only to them, not to colleagues, managers, or administrators.
Can we sign in with our own identity provider?
Yes. SSO and MFA are supported, and every access is recorded in a complete audit trail, so when an auditor asks who saw what, you can show them.
See Enterprise Security on your own kind of data
One conversation, a demo environment built on mock data modeled on your organization, and two weeks inside it to decide.
