Agentic FabriqMaintain an inventory of agents, owners, connected systems, and granted capabilities.
Every registered agent shows up in one view with a named owner, so “what is this bot and who runs it” stops being an investigation.
SOLUTIONS / IT
Know every agent you run — and control what it touches
Run the agent fleet from one place: what exists, who owns it, what it connects to, and how to shut it off.
01 / IDENTITY02 / POLICY03 / AUDIT
01 / THE CONTEXTEvery department is
Every department is
shipping agents.
As departments deploy their own copilots and agents, IT inherits a new class of access problem. Agents connect to SaaS applications, databases, internal APIs, and automation platforms—often through individually configured service accounts, OAuth grants, and API keys.
| Agent | Owner | Connects to | State |
|---|---|---|---|
| support-triage | Support ops |   Zendesk, Slack | Running |
| repo-assistant | Engineering |   GitHub, Jira | Running |
| deal-desk | Revenue ops |   Salesforce, Gmail | Running |
| deck-drafter | Marketing |  Google Drive | Disabled |
One row per registered agent: its named owner, the tools it may reach, and whether it is allowed to act at all. Disabling a row stops that agent’s access everywhere it connects, without changing a single downstream app.
02 / IN PRACTICE
Autonomy, with accountability.
ILLUSTRATIVE · ONE CALL, FOUR CHECKPOINTSit-agent, acting for an IT administrator: “read the assigned service inventory”
- 01Registered?no
In the inventory, with a named owner and a stated purpose.
Blocked · no credential, no routeA browser copilot nobody registered asks Slack for a private channel’s history. - 02Still switched on?no
The agent and the connection it would use are both enabled.
Blocked · the kill switch is here, not in the appdeck-drafter was disabled at 08:40; its 08:41 call arrives anyway. - 03Inside both grants?no
The agent’s scope and this administrator’s own permissions.
Blocked · the smaller of the two winsIt asks for another team’s project, which the administrator cannot open either. - 04In the agent’s own grant?no
Read was granted to this agent; change was not.
Not offered · the tool never reaches the agent’s list
Change a production configuration on the monitored service. - ThroughThe call runs.
Allowed · brokered credential, recorded with both identitiesRead the assigned service inventory.
Only one path reaches the tool, and every branch that leaves the spine is written down as well — so “why did nothing happen?” has an answer that does not start with reading an agent’s code.
03 / WHAT CHANGES
Control, in the details.
Standardize how agents sign in to internal and third-party services.
Connections run on credentials Fabriq holds and manages, instead of per-project API keys and one-off grants nobody tracks.
Replace unmanaged shared credentials with managed connections.
Existing service-account sprawl collapses into scoped connections that can be inspected, rotated, and revoked from one place.
Automatically constrain agent access when a user changes roles or leaves.
Fabriq inherits the change from your identity provider and narrows or removes the authority that user had delegated to agents.
Disable an agent or connection without changing every downstream application.
A kill switch in one place: one action stops the agent’s access everywhere it connects, with no emergency redeploys.
Review access centrally instead of inspecting individual agent codebases.
Access reviews read from one central record of grants and activity — no digging through a dozen codebases.
Apply consistent controls across cloud, SaaS, on-premises, and custom systems.
The same policy vocabulary covers a SaaS API, an internal database, and a legacy system, because enforcement sits in front of all of them.
THREE LAYERS. ONE DECISION.
The boundary travels with the work.
Explore the context attached to every request.
What is this agent allowed to do?
it-agent has a defined purpose and a bounded set of tools.
Whose authority travels with it?
The request acts for it administrator, using that person’s permissions.
Where does the action land?
The policy applies to the requested action within Service directory, before the tool executes.
04 / WAYS TO PUT IT TO WORK
Built for your real workflows.
01Offboarding an employee and removing the authority inherited by their personal agents.
Disabling the human account also revokes the grants their personal agents inherited — the access path IAM offboarding usually misses.
Identify→Authorize→Record
02Reviewing every agent connected to Google Workspace, Slack, Salesforce, or an internal database.
Filter the inventory by connected system to see every agent touching it, who owns each one, and what it can do there.
Identify→Authorize→Record
03Rotating or revoking a credential without redeploying the agent.
Credentials live in the broker, not the agent, so rotation and revocation happen centrally with zero downtime for the workflow.
Identify→Authorize→Record
04Making IT the only route into a new enterprise system.
An agent reaches a system only once IT has connected it and granted the scope. Nothing attaches on its own, so a shadow integration has no credential and no endpoint that answers it.
Identify→Authorize→Record
QUESTIONS
A closer look.
Does Fabriq replace our identity provider or endpoint tooling?+
No. It drops into the stack you already run and inherits users, groups, and role changes from your IdP. It adds the layer those tools do not model: agent identities, per-user delegated access, and action-level grants for every system agents touch.
Can we shut down a misbehaving agent quickly?+
Yes. Disabling the agent or its connection at the gateway stops its access to every system it reaches in one action, without redeploying applications or hunting down credentials.
What happens when an employee leaves?+
Their sessions revoke immediately and their agents lose access with them — while the record of everything those agents did stays intact for review.
Can we decide what each employee’s agents can reach?+
Yes. Access resolves per user: an agent connected by someone in support reaches support systems within that person’s permissions, and nothing else. One permissions layer covers every team’s agents instead of per-application configuration.
How long does rollout take?+
Sign-in comes through your existing identity provider, so there are no new accounts to create — connect your tools, register the agents people already use, and grants apply on the next call.
Does this cover the AI tools we haven’t approved?+
Yes, by architecture: keep tool credentials in Fabriq’s vault and unregistered agents have no key, no route, and no way in — nothing to hunt.