AI knowledge assistants

AI Knowledge Assistant & Copilot Development

Salt builds custom AI knowledge assistants and copilots that retrieve approved content and provide source-linked answers. Help employees, support agents, and sales teams find guidance, draft responses, and take explicitly approved workflow actions with access controls and human escalation.

Example use cases

What can we build with ai knowledge assistants?

These are examples, not a complete list of what we can build. We develop custom solutions around your business process, including use cases not listed here. Delivered client work is identified separately below.

01Internal operations

Employee policy & SOP assistant

Employees cannot quickly find the current approved procedure.

Start with
Approved policies, SOPs, document versions, and user permissions.
What we can build
Retrieve relevant passages and return an answer with links to the source.
Your team gets
A sourced answer or a clear escalation when the material is insufficient.

Plan for: Name a content owner and remove superseded policies from the active collection.

Discuss this use case
02Support teams

Customer support agent assistant

Agents switch between manuals and resolved tickets to answer a question.

Start with
Approved help articles, product manuals, and curated resolution notes.
What we can build
Suggest an answer and supporting sources for the support agent to review.
Your team gets
A draft response that stays under the agent’s control.

Plan for: Exclude private customer records unless the access model explicitly permits retrieval.

Discuss this use case
03B2B sales teams

Sales enablement assistant

Representatives need current product facts for proposals and buyer questions.

Start with
Approved product specifications, sales collateral, and commercial policies.
What we can build
Find relevant material and draft a sourced response to a product question.
Your team gets
A fact-checkable starting point for a sales reply or proposal.

Plan for: Prevent the assistant from inventing pricing, contractual terms, or unsupported capabilities.

Discuss this use case
04Engineering and product support

Technical documentation assistant

Integration guidance is spread across versions and repositories.

Start with
Versioned API documentation, runbooks, and approved troubleshooting guides.
What we can build
Retrieve guidance for the selected product version and link to the relevant instructions.
Your team gets
A version-aware explanation with the source available for verification.

Plan for: Test version conflicts and outdated guidance in the evaluation set.

Discuss this use case

Don’t see your use case? We can build around it.

Your project can adapt these examples, combine several workflows, or solve a different problem. Tell us what you need to achieve and which systems you use. We’ll define the scope, integrations, and a practical first release with you.

Discuss your workflow

The solution in action

From a daily task
to a connected workflow.

An illustrative flow to make the scope tangible. We adapt the steps, roles, and controls to your business.

  1. 01

    Ask a work question

    Search the approved content available to that user.

  2. 02

    Review a sourced answer

    Show supporting passages and document links.

  3. 03

    Escalate uncertainty

    Route missing or conflicting information to a person.

What we can build

The capabilities behind the workflow.

Source-connected answers

Retrieve relevant material before generating a response.

Permission-aware retrieval

Apply access boundaries to search and answer generation.

Evaluation and feedback

Test known questions, missing answers, and unsafe disclosures.

Controlled workflow actions

Connect approved APIs for actions such as creating a draft request, with permission checks, explicit confirmation, and an audit record.

Protocol knowledge-assistant project

KnowledgeBot protocol assistant

Salt’s KnowledgeBot project for Addiction Doctors focused on a custom AI agent that helps nurses and clinicians access medical protocols.

Explore what Salt delivered
Inside the project
  • Custom AI agent
  • Medical protocol information access
  • Nurse and clinician users

Read the project scope and implementation details to assess the fit for your business.

Plan before you build

Agree on the inputs.
Know what success means.

Systems, data & access

An approved content inventory, document owners, permission rules, model access, and a representative question set. For workflow actions, define permitted APIs, confirmation rules, and failure recovery.

What to measure

Answer correctness on reviewed questions, citation quality, escalation rate, and cost per interaction.

A focused first release

One knowledge collection, a defined user group, and a reviewed evaluation set.

Delivery with SPARK

Make the next decision with evidence.

Explore the SPARK framework

Define the workflow, test the uncertain parts, and release an agreed scope. Use the pilot to decide what belongs in the next stage.

  1. Agree on the scope. Map users, exceptions, systems, and acceptance criteria.
  2. Test the first workflow. One knowledge collection, a defined user group, and a reviewed evaluation set.
  3. Build and validate. Review working software, access boundaries, and integration failures.
  4. Release and improve. Establish a baseline using the agreed measures and review what users need next.

Before you decide

Questions buyers ask.

Connected expertise

Explore your next step.

AI & Automation

Let’s define a useful first release

Tell us where the work gets stuck.

Bring your current workflow, the systems involved, and the outcome your team needs. We’ll help identify a practical starting scope.

Discuss your workflow