tekBlueprint

The latest version of tekBlueprint.infra introduces an AI assistant (agent) subsystem that helps fill out hardware and software forms using natural language. The assistant can search the component database, look up specifications online, and see the context of your current project — all accessible through a chat sidebar directly in the editing interface.

Introduction

IT infrastructure planning involves working with dozens of parameters: CPU core count, memory capacity, storage types, network bandwidth, backup requirements, and much more. Every infrastructure element requires precise data, and manual entry is time-consuming and error-prone.

In this article, I’ll show how the AI assistant in tekBlueprint.infra solves this problem: it understands project context, knows the schemas of all entities, and can generate correct data from a single natural language phrase.

How It Works

Architecture

The AI assistant is built on one of the frameworks for working with large language models. Inference runs through an OpenAI-compatible API with streaming token support. Data from the model reaches the client in real time, allowing you to see responses as they’re generated rather than waiting for completion.

On the backend, each assistant is a task-specific agent with its own instructions, role, tools, output format, and constraints. For example, the server assistant knows the full schema and all valid enumeration values (server types, CPU architectures, memory formats).

15 Specialized Assistants

Every entity in tekBlueprint.infra has its own AI assistant:

Components: Hardware (6 assistants):

• Server — server creation and configuration.
• CPU — processor selection by architecture, core count, and clock speed.
• GPU — graphics card selection for compute workloads.
• Memory — calculation of memory capacity and type.
• Storage — storage selection by type, interface, and performance (IOPS).
• Network — network port and bandwidth configuration.

Components: Software (1 assistant):

• Software Service — services from the component library.

Project-level (8 assistants):

• Project — general project information.
• Security Zone — security zones.
• Environment — environments (dev, test, staging, prod, etc.).
• Software Service — services with resource requirements (CPU, RAM, storage, network, GPU).
• Software System — systems with purpose descriptions.
• Software Flow / Software Generator / Stream Redirect — data flows between modules.
• Software Backup — backup configuration.

Assistant Tools

The assistant doesn’t just generate text — it can perform actions:

Component Search — searches the hardware database by query. For example, “find an HP ProLiant DL360 Gen10 server with 8 SFF drives.” The assistant uses this tool to select correct templates and avoids inventing non-existent components. Integrated with tekBlueprint.kbase.

Web Search — allows the assistant to look up technical specifications and instructions on the internet. Useful when working with new hardware or technologies not in the local database.

Project Context — a set of 13+ tools that give the assistant visibility into the current project: lists of systems, services, modules, data flows, backups, environments, and security zones. This enables generating data consistent with the existing model.

Integration with Edit Forms

The key difference between the AI assistant and a regular chatbot is direct integration with edit forms. When you open a form to create or edit a server, service, or any other entity, an assistant button appears nearby.

Workflow

1. You start filling out the form manually (or leave it empty).
2. Click the AI Assistant button — the chat sidebar opens.
3. Describe your requirements in natural language: “server for PostgreSQL database, 64 GB RAM, NVMe SSD.”
4. The assistant generates a response and displays it as a structured block with form fields.
5. If everything looks correct, click “Apply” — the data populates the form.

If you’ve already partially filled out the form before calling the assistant, the current values are sent as context, and the AI takes them into account during generation. This lets you use the assistant to refine an existing configuration, not just create one from scratch.

❮ ❯

×

Output Format

For hardware, the assistant generates descriptions according to the required schema; for software — aligned with form fields. This allows applying changes selectively in the desired format while supporting established data representation conventions. For example: CPU cores, GiB of memory, Gbit/s network, storage IOPS, GPU TFLOPS.

What the User Sees in Chat

The AI assistant chat shows not only the final response but also the entire generation process:

Assistant Response — the main response text with a copy button. Convenient for saving snippets or transferring to documentation.

Assistant Reasoning — a specialized block showing the model’s reasoning process. Lets you understand decision-making logic and verify its correctness.

Tool Calls — service tool invocations with arguments. You can see which tools the assistant used and with what parameters: component search, web request, or project context query.

Statistics — information about token consumption and response time. Useful for estimating request costs and performance.

Execution Control

AI inference is an external service that may be unavailable, slow, or produce incorrect results. To protect the system from LLM endpoint issues, several mechanisms are used:

• Concurrency Limits — maximum simultaneous calls to the model. This is necessary to comply with provider-side rate limits and prevents server lockups or resource exhaustion under peak load.
• Request Rate Limits — configurable quotas on the number of requests per time period. Protects against abuse and ensures fair resource distribution among users.
• Data Filtering and Validation — data received from the model is validated and mapped to internal representations. This prevents errors during auto-fill or incorrect model responses.
• Role-based Restrictions — assistants have constraints depending on the user’s role.

Additional Improvements

Beyond the AI assistant, this release includes other changes:

Enhanced Server Metrics — extended load metric processing with a new dialog for adding services to servers. More pre-configured metrics are now supported (CPU, data, IOPS, memory, network (rx), network (tx), RPS, store, VRAM) and displayed in a unified style.

Local Component Repository — an alternative to using the external tekBlueprint.kbase. An internal provider of customizable component templates from the database.

Markdown Editor — descriptions in edit forms are gradually being migrated to Markdown instead of plain text fields, improving the visual presentation of generated documentation.

Conclusion

The AI assistant in tekBlueprint.infra is not just a chatbot — it’s a working tool for infrastructure design. It understands project context, knows the schemas of all entities, can search components in the database and look up specifications online, and its generated results are applied directly to edit forms with a single click.