AI Chat and RAG Features

Akurion chat combines approved project knowledge, metadata, relationship context, and answer generation. The goal is not only to produce a fluent answer, but to produce an answer users can inspect, verify, and reuse.

Answer Modes

Mode	Best for	Behavior
Fast	Simple questions and low-latency answers	Retrieves relevant chunks and streams a direct cited answer.
Research	Broader analysis	Uses deeper retrieval and synthesis for more complete answers.
Agentic	Multi-step reasoning	Plans retrieval, searches multiple ways, and reasons over intermediate evidence.
MCP-enhanced	Tool-driven advanced RAG	Uses Akurion’s advanced knowledge tools to support deeper research, stronger source grounding, and clearer citations.

How Answers Work

At a business level, Akurion helps users move from a question to a verifiable answer by:

Understanding the user’s question in the context of the current project.
Looking across the approved knowledge base and any selected filters.
Prioritizing the most relevant source material for the task.
Using graph context when relationship discovery is enabled.
Producing an answer with citations and source details that users can verify.
Recording usage information for billing, analytics, and governance.

Search Modes

Advanced modes use search strategies such as:

Search mode	Use case
`focused`	Direct answer over a defined topic.
`exploratory`	Broad discovery when the user does not know the available content.
`comparative`	Compare documents, sources, vendors, customers, or time periods.
`timeline`	Find chronological changes or events.
`relationship`	Explore relationships between entities, systems, people, or records.

Metadata Filters

Metadata filters narrow retrieval to relevant content. They can be generated from natural language or selected in the UI.

Examples:

“Show 2025 legal documents for EMEA.”
“Find policies owned by Security.”
“Compare incident reports for application X last quarter.”
“Summarize customer-facing support articles for product Y.”

Developer APIs accept structured metadata filters:

{
  "department": { "$eq": "legal" },
  "year": { "$gte": 2025 },
  "region": { "$in": ["EMEA", "APAC"] }
}

Citations and Sources

Akurion answers include source metadata when available:

File name or source record name.
Page or chunk reference.
Source type.
Original URL.
Rendering or preview support.
Excerpt text.
File ID for follow-up API calls.

Use citations to verify answers and open the original source.

Graph RAG

Graph RAG adds relationship-aware context to retrieval. When enabled, Akurion can extract entities and relationships from processed files, store them as graph records, and use them during answers.

Graph RAG helps with:

Entity-centric questions.
Relationship questions.
Impact analysis.
Cross-document discovery.
Duplicate or alias-heavy corpora.
Questions where the answer depends on how records connect.

Web Search

Projects can enable or disable web search. Use web search only when external context is allowed for the project.

Good uses:

Compare internal information with current public context.
Fill public background context around internal facts.
Research market or technical topics alongside private docs.

Avoid web search for:

Strictly confidential projects.
Tasks that must only use approved internal sources.
Regulated workflows where external context is not allowed.

Prompt Playbooks

Prompt playbooks help teams standardize common tasks:

Summarize a source set.
Compare documents.
Extract obligations.
Create executive briefs.
Draft FAQs.
Analyze incidents.
Prepare account research.

Use playbooks when a team repeats the same analysis pattern.

Feedback and Quality

Use answer feedback to improve project quality:

Helpful or not helpful.
Missing source.
Wrong citation.
Outdated content.
Needs more detail.
Unsafe or inappropriate answer.

If an answer is weak, improve the project by adding better sources, adding metadata keys, adjusting project instructions, or using a deeper search mode.