/
/
/
Česky
A platform to define, run, and govern AI agents in production. Many agents per project, each with versioned instructions, its own model and tools, and triggers from a button, a schedule, or a webhook. Every run is recorded — full transcript, streamed events, tool calls, token usage, and cost. Built on .NET and the Microsoft Agent Framework, running in your own Azure tenant behind Entra ID.
Book a demoMost agent demos stop at the chat box. Production starts when the same agent has to run unattended, on a schedule, against your real systems — with someone able to read back exactly what it did and what it cost. That's the part we built.
An agent that triages inbound requests every few minutes, pulls context from your CRM over MCP, and drafts a reply for someone to approve.
A scheduled agent that reconciles yesterday's invoices against the ERP overnight and files the exceptions for review.
A webhook-triggered agent that reviews every pull request, runs your checks, and posts a summary back to the team.
Group agents by project. Each project carries its own catalog of model providers — Azure OpenAI, OpenAI, Foundry — and MCP servers, shared across the agents inside it.
Every system prompt is versioned, with full history and a diff view. Change an agent's behaviour deliberately, see exactly what changed, and roll back when a change misfires.
Bind MCP servers — stdio or HTTP/SSE — at the project or the agent level. Agents get structured access to your systems; enable, disable, or override the binding per agent.
Run an agent by hand, on a cron schedule, or from a webhook secured with a shared secret. The same agent fits an ad-hoc task and an unattended pipeline.
Every run keeps the full transcript, streamed events, and tool calls, with token usage and cost — a live estimate during the run plus reconciled actuals afterwards.
Cost rolls up per project. Retention is configured per project — separately for events and transcripts — so governance is wired in from the start, not bolted on for the audit.
Anyone can wire a model to a prompt in an afternoon. The cost shows up later: when the prompt changes and nobody knows what it used to say, when the agent runs unattended and fails silently, when finance asks what it spent, when an auditor asks what it did with that data.
The platform answers all four by default — versioned instructions, recorded runs, per-project cost, and retention you control. That's the difference between a clever demo and something you can run in production for years.
.NET, the Microsoft Agent Framework, Havit.Blazor for the UI, SQL Server or Azure SQL for storage, and Microsoft Entra ID as the only way in — every page and API is authorized. No data leaves your environment for a third-party SaaS.
It deploys into your Azure tenant — App Service, Azure SQL, Key Vault, Application Insights — and hosted MCP servers run as isolated Container Apps behind a shared-secret proxy. Your agents, your tenant, your data.