Agentic FabriqGive product and internal agents distinct identities.
Product copilots and internal automations stop sharing service accounts; each is a principal you can reason about and revoke.
SOLUTIONS / SaaS / Technology
Build agent capabilities customers can trust
Add governed agent identity and permissions to products, internal tools, and developer workflows.
01 / IDENTITY02 / POLICY03 / AUDIT
01 / THE CONTEXTShip for every tenant
Ship for every tenant
without reaching across them.
Technology companies are adding agents to products while also deploying internal agents for engineering, support, sales, and operations. Product teams must determine how an agent acts within each customer tenant, on behalf of each user, and across connected applications.
02 / IN PRACTICE
Autonomy, with accountability.
THREE TENANTS, ONE AGENT, NO DOOR BETWEEN THEMIllustrative · customer-agent, acting for a signed-in customer user
THIS SESSIONTenant · AcmeThe tenant the signed-in user belongs to
- Read the records this user owns
LinearAllowed. Tenant and user are both resolved from the session, not from the prompt. - Read this tenant’s error events
SentryAllowed, and scoped to this tenant’s own events — the query cannot widen. - Change tenant configuration
GitHubHeld. Configuration belongs to an admin of this tenant, so the change waits for one.TENANT ADMIN
Tenant · BorealisSame product, same agent, another customer
- Read another tenant’s dataSentryBlocked at the boundary. Fabriq refuses it before a call is made, so Borealis’ systems never see the request.
Tenant · CorvusA tenant this user has never signed in toNo grant, no session, no route — the same wall, drawn once and applied to every tenant the same way.
Example tenants. The boundaries, roles and grants are the ones you set.
03 / WHAT CHANGES
Control, in the details.
Enforce tenant, workspace, user, and resource boundaries.
Boundaries are enforced at the tool-call layer, so a multi-tenant product agent cannot cross tenants even if application logic slips.
Apply the customer’s own permissions to agent actions.
The agent acting for a user inherits that user’s permissions inside their tenant — no more, even when asked.
Let each customer set the grant your agent runs under in their tenant.
Enterprise customers choose which actions your agent may take for each of their users, and the agent is capped at the smaller of that grant and the user’s own permissions.
Leave the destructive and outward-facing actions out of the grant.
An action your customer did not grant never reaches the tool list your agent is handed, so the step stays in a person’s hands inside your product’s own experience.
Control third-party integrations and the credentials behind them.
Customer-connected integrations run on credentials Fabriq holds and limits, so your product never stores them raw.
Provide enterprise customers with clear agent activity and audit evidence.
Per-tenant activity logs give your customers the audit story their security teams demand.
Remain independent of the underlying model and agent framework.
Swap models or frameworks without re-answering the enterprise security questionnaire — controls live in the layer, not the stack.
THREE LAYERS. ONE DECISION.
The boundary travels with the work.
Explore the context attached to every request.
What is this agent allowed to do?
customer-agent has a defined purpose and a bounded set of tools.
Whose authority travels with it?
The request acts for customer user, using that person’s permissions.
Where does the action land?
The policy applies to the requested action within Tenant workspace, before the tool executes.
04 / WAYS TO PUT IT TO WORK
Built for your real workflows.
01Product copilots
Copilots act within each tenant under the end user’s own permissions, and reach only the actions that tenant granted.
Identify→Authorize→Record
02Coding agents
Engineering agents are granted branches and pull requests; merge and deploy are not in the list they receive, so those stay with an engineer.
Identify→Authorize→Record
03Support automation
Support agents resolve within the customer’s account using the actions the rep they assist already holds; anything past that is refused and lands with a person.
Identify→Authorize→Record
04DevOps agents
Operations agents read production context freely; changing infrastructure is not an action they hold, so the on-call engineer makes it.
Identify→Authorize→Record
05Customer integrations
Customer-connected apps run on credentials Fabriq holds, rotates, and revokes centrally.
Identify→Authorize→Record
06Internal knowledge assistants
Assistants answer from approved sources under each employee’s permissions — internal data stays limited by role.
Identify→Authorize→Record
QUESTIONS
A closer look.
How does Fabriq drop into our product architecture?+
As a layer between your agents and tenant data — not a rearchitecture. Your product keeps its stack, model, and framework; agent tool calls route through Fabriq, which resolves tenant, workspace, and user boundaries on every request.
How does tenant isolation work for product agents?+
Isolation is enforced at the tool-call layer with the tenant, workspace, and acting user in every authorization decision. Even a bug or manipulated prompt in the agent cannot cross tenants, because the boundary sits outside the model and the application code.
How do our customers revoke access?+
Disconnecting kills the connection’s token immediately, and session revocation takes effect now — per user, without touching other tenants.
Can our customers set per-user, per-tenant policy?+
Yes. Each enterprise tenant sets its own grant for your agent, and the agent is capped at the intersection of that grant and the acting end user’s own permissions inside the tenant — your customers choose what agents can access and do for each of their users.
What do we get out of the box versus build?+
You provision users and call the API; Fabriq handles per-user OAuth, token storage, scoping, and the audit trail underneath your product.
What does the audit story look like for our buyers?+
Every agent action in your product carries the end-user identity and the verdict — evidence your enterprise customers can query when they ask what the agent did.