Skip to content

VegaDūta documentation

These pages are also published as a searchable site (mkdocs.yml, make docs-serve); see Publishing these docs.

Start here

Page For
Getting started Lite install (installer or checkout), first agent, first workflow
Installation Personal (Setup app or CLI), Lite, Full (Docker Compose) and Kubernetes (Helm); images and signature verification
Console guide Every signed-in screen: the Assistant surface, build, channels, monitor, admin
Configuration Spring profiles, core shapes, environment variables, feature flags
Extending Tools from OpenAPI, MCP servers, SKILL.md skills, LLM providers, workflow nodes (NodeHandler, NODE_FORMS), channels (adapter registry), tool sources, the HTTP request pipeline, LLM call sinks, the turn journal
SDKs and dependencies The Java and Python Agent SDKs (governance inside your own application, plus the Java risk scanner) and the generated TypeScript and Python API clients
Plugins, apps and mobile Editor and browser plugins, the Android app, the MCP server: status, catalog, publishing
Repository structure How the repository is arranged and how the platform, public pages, SDKs, docs and plugins are linked
Site modes and public pages Open-source vs commercial public pages, and where every surface is linked
Architecture overview Deployables, data, authentication, and why the chat path stays in one process

Features

Page For
Personal mode One person on one machine: the Setup app (installs Docker and a free local model, repairs a failed start) or scripts/vegaduta personal, local sign-in, local models, the Assistant's workspace and desktop tools, skills, sandbox levels, limitations
Cost and ROI Per-agent cost, outcome rates, tenant-entered value assumptions; /roi and /api/tenant/roi/**
Running a 30-day pilot, case-study template The pilot process, baseline, one-page pilot report, demo data (scripts/vegaduta seed-demo)
Run free with local models Ollama in Compose or Helm, the model catalog (scripts/vegaduta models), on-device models at /edge
Share an agent as a file Export an agent as a bundle, import one safely, ready-made examples in examples/agent-bundles/
Edge relay A signed-in browser as the model behind your SMS and WhatsApp numbers
Channels overview All fifteen channels in one table (how each connects, its endpoint, what you provide), the rules they share, and how each inbound call is authenticated; one page per channel
SMS conversations Plain-text channel on a Twilio number
Signal, Matrix, iMessage Personal channels through a signal-cli server, a Matrix bot user and a BlueBubbles Mac; default deny with pairing
WhatsApp QR Link your own WhatsApp number as a companion device — no Meta Business account, no public URL
WhatsApp policy Meta's rules as read on 2026-10-14 and what the policy guard enforces
DPDP obligations and controls Consent, purposes, retention, export and erasure; engineering notes, not legal advice

Deploying

Reference

Operating

Maintainers

Project