Use cases / Knowledge and engineering
AI Research
Turn scattered sources into cited, reviewable research.
01
Problem
Analysts spend most of their time finding and reconciling sources. Generic AI tools cite nothing, mix public and confidential material and leave no record of what was read.
02
AI capability
Plan a research question, search approved sources, read and compare documents, and produce structured findings with citations.
03
Data
The agent works from the context you connect, not from the open internet by default.
- Internal reports and notes
- Licensed or approved external sources
- Data room and shared drives (permission-scoped)
04
Agent
Research Agent
Runs multi-step research and returns cited briefs for an analyst to review.
05
Workflow
- 1. Define question
- 2. Select permitted sources
- 3. Search and read
- 4. Cross-check claims
- 5. Draft cited brief
- 6. Analyst review
- 7. Publish to workspace
06
Governance
Governance runs inside the agent at runtime: policy changes what it can actually do.
Can
- Read sources the requester is already permitted to see
- Search approved external sources
- Summarize and compare with citations
- Save drafts to a workspace
Cannot
- Send confidential material to unapproved models or tools
- Access restricted folders
- Publish externally
- Present uncited claims as findings
Requires approval
- Use of a new external source
- Distribution outside the team
Records
Agent identity, Data accessed, Model used, Output, Tools called, Policy applied, Decision, Approval, Action, Outcome.
Suggested starting autonomy: L1 Assist
Start at L1 where the agent recommends and the analyst decides; research output is reviewed before use by design.
07
Outcome
What you measure, so the agent earns more autonomy on evidence:
- Time from question to reviewed brief
- Share of claims with a verifiable citation
- Analyst correction rate
- Source coverage
Questions
Does research data leave our environment?
The platform is designed so approved models and data residency are set per Trust Profile. Sources and models outside the approved set are denied.
How do we check the agent was right?
Every finding links to the passages used, and the run record shows sources read and model used.