Search

What are you looking for?

Enter at least 3 characters.

AI in AMODIT Document Workflows: Copilot, AskAI, RAG and Architecture

Running an AI model is only one part of a controlled document workflow. The model also needs the right instruction, case context, organizational knowledge and approved tools. This guide explains how AMODIT Copilot, AskAI and the RAG Knowledge Base fit into that architecture, and where the knowledge base and model can operate.

Copilot, AskAI and a RAG knowledge base connected to document workflow and human review

1. Where Does AI Support Document Workflows?

In enterprise, large and mid-sized organizations, one document can move across multiple roles, systems and control points. AMODIT uses AI where work requires understanding content, finding the right organizational knowledge or preparing a proposed next step.

  • Copilot helps users navigate AMODIT, identify urgent work and prepare a report on request.
  • AskAI works with the open case and its documents.
  • The RAG Knowledge Base adds organizational context from approved procedures, instructions and policies.
  • Agent rules pass an instruction and process context to AI, then return the output to the configured workflow.

AI recommends; people approve. Users review outputs whenever the process requires assessment, acceptance or a business decision.

2. How Do Copilot, AskAI and the RAG Knowledge Base Differ?

  • AMODIT Copilot supports users and process designers. It can assist with navigation, reports, process logic and drafts of workflows, rules or documentation.
  • AskAI analyzes the current case, including form data, fields, comments and attached documents. It can summarize content, surface important information or inconsistencies and answer a question about that case.
  • The RAG Knowledge Base supplies controlled organizational context that is not contained in the model or a single prompt.

An implementation uses existing models and configures them for the organization through model selection, instructions, knowledge, context and integrations.

3. What Makes Up the AMODIT AI Architecture?

An AI response or action is built from several connected elements:

  1. Model: analyzes the supplied data and prepares an output.
  2. Instruction: defines the task, expected response and constraints.
  3. Context: supplies data from the relevant case, document or workflow step.
  4. Knowledge: adds approved organizational material.
  5. Tools and actions: define which exposed capabilities an agent may use and how its output returns to the workflow.
  6. Output: reaches the user or the next configured process step.
  7. Logs and limits: let administrators observe executions, LLM conversations, usage and available limits, and support diagnostics.

Administrators configure the instruction, context, knowledge, tools and use of the output. The model is one component of the solution, not an independent process owner.

4. Where Can the Knowledge Base and Model Run?

The AMODIT AI architecture includes four options. In options A–C, the model analyzes data in the cloud. In option D, it runs in the customer environment.

OptionKnowledge BaseModel and Flow
A — Knowledge Base at AstrafoxHosted in the Astrafox environment.The Astrafox AI Gateway sends requests to a Microsoft or Google model using an Astrafox account and returns the response. The customer does not need its own cloud account.
B — Knowledge Base at the CustomerHosted in the customer environment.The AI Gateway remains at Astrafox, while a Microsoft or Google model runs through an Astrafox account. The customer keeps the Knowledge Base locally without maintaining its own cloud account. This option is marked as recommended.
C — Customer Cloud AccountHosted in the customer environment.The model runs through the customer's Microsoft or Google account. The customer provides its own Azure or Google account.
D — Customer-Hosted Local ModelHosted in the customer environment, with knowledge vectors created locally.The model runs locally, so the Knowledge Base and data analysis remain in the customer infrastructure.

Selecting an option is part of designing the architecture for a specific environment. The option overview does not define security parameters, retention or cost.

5. What Can AMODIT Copilot Do?

Copilot supports daily user work and process design. It can help users:

  • navigate AMODIT and identify work that needs attention,
  • prepare a report or summary on request,
  • draft a workflow with steps, forms and fields,
  • propose rule code, explain it or help improve it,
  • prepare a process documentation draft from the current configuration.

For example, a process owner can request a contract workflow with intake, review, approval and archiving stages. The owner reviews the draft and refines the configuration before release.

6. What Does AskAI Do?

AskAI works in the context of the open case. It uses form data, fields, comments and attached documents to answer a question about that specific piece of work.

A user might ask: “Summarize this agreement in five points and identify the key obligations.” The response is material for employee review, not a decision made outside the workflow.

7. How Does the RAG Knowledge Base Differ from AskAI?

CapabilitySource of ContextPurpose
AskAIThe current case and its attached documents.Understand, summarize and analyze content in a specific process.
RAG Knowledge BaseApproved procedures, instructions, policies and other organizational materials.Answer questions using broader organizational knowledge.

Administrators maintain knowledge sources, freshness, agent connections, access and constraints.

AMODIT also provides a separate area for files that do not belong to a case or process. Read A File Repository Outside the Case Workflow.

8. How Do Administrators Control AI Operations?

Administrators manage agents built for specific tasks and can define a custom Copilot agent. Configuration covers instructions, context sources, knowledge, tools, access and constraints.

Operational views cover LLM logs, LLM conversations, usage statistics, available limits, analysis and diagnostics. They provide evidence for checking an execution and for an administrator's follow-up investigation.

9. What Must Be Confirmed for a Specific Environment?

The following parameters remain environment-specific:

  • feature availability by AMODIT version or license,
  • supported model names and versions,
  • retention for prompts, responses, logs and documents,
  • detailed security controls and data-processing terms for each architecture option,
  • licensing, setup or model-usage costs,
  • limit values or how those limits are calculated.

These parameters need confirmation for the selected environment, model and implementation scope. The functional architecture explains the role and location of each component; it does not replace technical and contractual arrangements.

Summary

Copilot, AskAI and the RAG Knowledge Base cover different parts of AI-assisted work in AMODIT. Their operation depends on the complete architecture: model, instruction, context, knowledge, tools, output, logs and limits. Four deployment options show where the Knowledge Base can reside and where the model analyzes data.

Discuss AI Architecture for Your Workflow

Article by:

Janusz Bossak, AMODIT Development Director

Janusz Bossak

AMODIT Development Director

Janusz combines technical knowledge with business experience in process optimization, performance improvement and work-cost reduction. He guides the development of the AMODIT platform. Photography is his passion.