Use AI and organizational knowledge
Move from first access to a governed AI workflow with models and approved content.
This path is for business users, analysts, subject-matter experts, and team leads who use BullSequana AI through the Portal rather than building or operating the platform.
Outcome
You can sign in, choose the right AI experience, work with an approved model or knowledge source, evaluate the result, and know where to go when the behavior is unexpected.
Before you begin
Confirm that you have:
- an account in the correct tenant
- membership in the team that owns the work
- access to an approved model or AI experience
- permission to use the source material involved
- a clear rule for what information may be entered or uploaded
If any of these are unclear, start with Access and authentication or contact your tenant administrator.
1. Understand the workspace
Read the Portal for sign-in and navigation, then use OmniChat for conversations, files, Agents, and common user actions. The Portal is the customer-facing experience; internal components are not separate user destinations.
OmniChat also offers:
- Build mode, where the agent runs code in its own sandbox and creates files or a web app you can preview, where your platform team has enabled it
- Deep Research for cited, multi-step research on a question
- Connectors, which let the assistant use approved external tools
- My usage, which shows your own consumption and your monthly cap; see Review your usage
2. Choose model-only or knowledge-grounded use
Use a model-only conversation for drafting, summarization, transformation, or general reasoning that does not depend on private organizational facts.
Use a knowledge-grounded experience when answers must use approved internal documents or data. RAG explains how content is ingested, indexed, retrieved, and added to a response.
For repeatable work, create an Agent with the intended sources or add a Skill to a text conversation. For a spoken exchange, use Voicebot; it does not use retrieval, Skills, or connected tools, so continue in text for knowledge-grounded tasks.
3. Protect the information boundary
Before uploading content or connecting a source, confirm its tenant, team, classification, owner, and retention expectations. The person who can start a conversation is not automatically allowed to use every source available to the organization.
Review Security and compliance when the use case involves regulated, personal, confidential, or jurisdiction-bound information.
4. Evaluate the result
Check whether the response:
- answers the actual question
- cites or reflects the intended source material
- distinguishes known facts from generated interpretation
- stays within the requested scope
- avoids disallowed or cross-tenant information
For repeatable business workflows, define examples of acceptable and unacceptable output before broad rollout.
5. Resolve common problems
Use Common questions for access, model, file, and retrieval behavior. Use Troubleshooting when a request fails, produces no retrieved context, or behaves differently across users.
A model you cannot pick has no allow rule for you; ask your tenant administrator for access. When your monthly BCU cap is reached, a banner says so and new requests are refused until a tenant administrator raises the cap or the month ends.
You are done when
- users can reach the intended experience with the expected tenant and team identity
- the model and knowledge sources are explicitly approved
- representative questions produce useful and permission-correct results
- users know the review and escalation path
- the service owner can observe adoption and recurring friction without exposing customer content
Next, use Build an AI application when the workflow must be embedded in another application.