AI agents

Tengri has built-in virtual assistants — AI agents controlled by LLM and possessing various useful skills. To use the help of an agent, just find its name in the chat list and start a dialogue with it by asking questions in natural language.

Key features of AI agents

Security

AI agents are strictly restricted in terms of access rights:

  • They have no rights to make changes to the data.

  • The user must explicitly grant them the rights to read data, just as is done for normal users.

Memory

AI agents have short-term and long-term memory to develop skills and engage in meaningful dialogues.

  • Short-term sessional memory: what are we talking about right now?

  • Long-term personal memory: how do we count this metric?

  • Long-term shared memory: what metrics/charts/tables are we using in our project?

Semantic layer

When working with data, AI agents use not only meta-data from schemas and tables, but also information from the semantic layer. These are detailed comments on schemas, tables, and columns that users can create on their own or have special agent create.

The semantic layer allows AI agents to better understand both the structure of the data and the questions that users ask about that data.

Through the use of the semantic layer, AI agents know the answers to questions such as:

  • What exactly do these terms mean in our project?

  • What are "customer", "revenue", MRR in relation to our project data or to a specific schema within the project?

  • What metrics are used in our project? How are they calculated?

  • On which datasets are our metrics calculated? In which columns of which tables is this data stored?

  • What attributes does our data have and what exactly do they mean (for each column of each table)?

MCP tools

AI agents' work with data is implemented using MCP:

  • MCP for SQL queries

  • MCP for working with memory

  • MCP for service development

Executing SQL queries inside a dialogue

The dialogue interface with AI agents in Tengri allows you to execute a real query against real data directly inside the dialogue and make sure that the AI agent has provided a correct query in response to the user’s question and not made up an answer.

AI agent Theodor: Analytics specialist

Theodor — an agent who can help you build SQL-query, analyse data, find patterns and insights.

Skills:

  • Understands the data question in natural language

  • Forms a SQL-query for the data as an answer

  • Provides the user with the generated query for self-execution and correctness checking

  • Interprets the answer

  • Draws conclusions

Examples of questions:

  • What is the average price of product X in period Y at company Z according to our invoices?

  • Derive the top 100 best selling seasonally adjusted products from the 2024 collection.

  • Find anomalies in the data from the tables in the sales chart.

Example Theodor cafe
Example of dialogue with agent Theodor

AI agent Dasha: Data visualisation specialist

Dasha — an agent that helps visualise data.

Skills:

  • Understands the data question in natural language

  • Forms SQL-query for data

  • Creates visualisations: dashboards, charts, graphs, diagrams

  • Automates visualisation in JavaScript

Examples of questions:

  • Construct a graph of the relationship between sales in units and the day of the week from the previous year

  • Visualise columns A and B from table X in chart Y

Example Dasha cbrf
Example of dialogue with agent Dasha

AI agent Archie: Data documentation specialist

Archie — an agent that deals with data description and semantic layer generation.

Skills:

  • Analyses table contents by metadata

  • Analyses the contents of tables by randomly selecting data from columns if it has read access to that data

  • Documents the data based on the information and real-world knowledge obtained

  • Forms a semantic layer to be used by other agents

Examples of questions:

  • Describe all the tables in schema X

  • Describe table Y

  • Supplement our descriptions for schema X with your own

Example Archie demo3
Example of dialogue with agent Archie