Agentic FabriqIsolate client and engagement data.
Each engagement is a sealed workspace; the same firm-wide agent cannot mix two clients’ materials.
SOLUTIONS / Consulting
Scale reusable agents without mixing client information
Separate client data and govern agents working across projects, deliverables, and delivery teams.
01 / IDENTITY02 / POLICY03 / AUDIT
01 / THE CONTEXTEach client engagement
Each client engagement
sealed from the next.
Consulting, accounting, and professional-services firms want to reuse agents across research, analysis, delivery, and knowledge management. But reusable agents can become a pathway between confidential client environments.
02 / IN PRACTICE
Autonomy, with accountability.
ONE QUESTION, TRACED END TO ENDIllustrative · project-agent, acting for a consultant on ENG-4410
- asked“Pull last week’s findings together for the steering pack”
SlackOne consultant, one engagement, one question typed in the client channel. - identityproject-agent, acting for this consultantBoth identities travel with the request: the agent’s, and the person’s own.
- scopeENG-4410 — resolved from the staffing record, not from the promptThe engagement a person is staffed on is what their agent can reach today.
- credentialWorkspace token injected server-sideThe key is used on the way out and never handed to the agent or written into a config.
- callRead the engagement’s own workspace folderallowed
Google DriveThree working files read and summarised, each cited back to where it came from. - callReach ENG-4527, a different client’s filesblockedGoogle DriveRefused inside Fabriq. No call leaves, so the other client’s workspace never sees a request at all.
- callSend the pack to an address outside the firmheld
GmailThis agent holds the draft action and not the send. The tool is not in the list it was handed, so the pack stays put and the partner sends it.NOT IN THE GRANT - recordedEvery line above kept as one recordWho asked, what acted, what it reached, what it was refused — one trail per request.
Example trace. The engagements, staffing and grants are the ones you set.
03 / WHAT CHANGES
Control, in the details.
Apply project-level access to shared agents.
Reusable agents resolve access per project at request time, so one deployment safely serves every team.
Prevent retrieval from unrelated client workspaces.
A broad research question stops at the engagement boundary instead of sweeping the firm’s document store.
Restrict subcontractors to approved engagements and time periods.
Subcontractor access is engagement-scoped and time-limited, ending when the statement of work does.
Decide whether an agent may deliver externally at all, or only draft.
Draft and send are separate actions, so an agent can prepare a deliverable, proposal or publication and have no route out of the firm.
Track which source materials contributed to each deliverable.
Every deliverable carries its sources, protecting both quality review and client confidentiality claims.
Revoke access automatically when a project ends.
Project close-out revokes the team’s and their agents’ access in one motion, with the record retained.
THREE LAYERS. ONE DECISION.
The boundary travels with the work.
Explore the context attached to every request.
What is this agent allowed to do?
project-agent has a defined purpose and a bounded set of tools.
Whose authority travels with it?
The request acts for project consultant, using that person’s permissions.
Where does the action land?
The policy applies to the requested action within Client workspace, before the tool executes.
04 / WAYS TO PUT IT TO WORK
Built for your real workflows.
01Client research
Research agents work within one client’s folder and cite sources, keeping parallel engagements separated.
Identify→Authorize→Record
02Document analysis
Analysis stays inside the engagement’s document set, however broad the question.
Identify→Authorize→Record
03Proposal creation
Proposal agents draw on approved firm collateral and the prospect’s own materials, and send only as the consultant they act for.
Identify→Authorize→Record
04Engagement delivery
Delivery agents assemble work product from project materials, and the client send goes out as the engagement lead — never beyond what that lead could do.
Identify→Authorize→Record
05Audit preparation
Preparation agents compile evidence within the audit engagement’s scope, preserving independence boundaries.
Identify→Authorize→Record
06Internal knowledge retrieval
Knowledge agents serve sanitized, approved precedents — client-confidential material stays in its engagement.
Identify→Authorize→Record
QUESTIONS
A closer look.
Do consultants have to change how they work?+
No. Fabriq drops in over the document stores and tools the firm already uses. Consultants and their agents keep working the same way; access simply resolves per person and per engagement instead of firm-wide.
What happens when an engagement ends?+
Close-out revokes the team’s and their agents’ access automatically, including subcontractors. The engagement record — what was accessed, produced, and delivered — is retained for quality review and confidentiality defense.
What can we show a client’s security team?+
SSO through your identity provider, credentials in a vault, default-deny permissions, and a full trail of what agents did on their account — answers, not promises.
Can one agent serve every team with per-person access?+
Yes — that is the model. The agent is deployed once, and each request is evaluated against the consultant’s current project assignments, so reuse never means shared data and staffing changes move the boundary automatically.
How is client confidentiality kept between engagements?+
Access is per person and per grant — agents see only the intersection of their scopes and each consultant’s own engagements, and every access is attributed.
Do consultants paste client credentials anywhere?+
No — connections are made once, per person, and credentials attach server-side at call time; laptops never hold the keys.