VegaDūta

Platform

The platform under your AI workforce

Agents are only as dependable as the platform they run on. VegaDūta pairs its agents with durable Temporal-backed workflows, a builder that works in plain language, real APIs, schema-per-tenant isolation, and tracing on every LLM call.

These pages document each platform capability as it is actually built — including what is not built yet.

Workflow Automation

AI workflows built from 35 node types, started from 11 trigger channels, with approval gates, data pinning, and draft runs.

Agent Builder

Describe the agent you want in plain language; the builder configures tools, channels, and guardrails.

Webhooks

Authenticated inbound webhook triggers for workflows, with signature-validated channel events underneath.

Developer APIs

A REST developer API with scoped keys and rate limits for calling your agents and workflows from code.

Security

Schema-per-tenant isolation, Keycloak SSO, encrypted credentials, guardrails, and audit logging.

Observability & Evals

Langfuse tracing over OpenTelemetry, per-agent quality scores, and sampled LLM-as-judge evaluations.

Agent Lab

A run graph you can fork from, prompt knockout to find which instructions matter, and a consistency check.

Agent Skills

SKILL.md packages agents load on demand: 14 built in, zip import and export, and admin review for agent-written drafts.

Agent Teams

A lead agent hands parts of a task to specialist agents, in parallel, inside fixed limits on branches, time, and spend.

Versions & Evals

Versioned agents and workflows with staging and production labels, eval datasets, six judges, and a promotion gate.

App Studio

Describe an app and get a multi-page build with live preview, Figma and GitHub import, and your agents wired in.

Governance

Tool confirmation, per-user and per-run budgets, an audit log with a live tail, and workspace SSO with SCIM.

Source-Available & Self-Hosting

BSL 1.1 licensed, with desktop installers, a Docker Compose stack and a Helm chart for running the whole platform on your own machines.

Try it before reading further

The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card.