Architecture

How It Works

Reva lives inside Microsoft Teams and connects to your release management and project tracking tools. Ask questions in natural language — Reva understands the intent, queries the right systems, and responds with a clear answer.

You & Your Team

Microsoft Teams

Chat with Reva in personal messages, group chats, or channels

Secured via Azure Bot Framework

Reva AI Engine

Intelligent Routing & Agent Loop

Understands your question, selects the right tools, gathers data, and composes a response — in the language you write in.

Model Context Protocol (MCP)

Your Systems

Digital.ai Release

Releases, pipelines, tasks, teams, gates

Jira Cloud

Issues, sprints, boards, projects

More Integrations

Jenkins, GitLab, ServiceNow, …

Infrastructure (on your premises)

Local LLM

AI inference runs entirely on your hardware

Database

Conversations, semantic & episodic memory

Directory Service

User & team resolution

From Question to Answer

When you send a message to Reva in Teams, it goes through four stages:

1

Understand Reva detects your language and classifies your question — is it about releases, Jira issues, or something else?

2

Gather The AI agent queries the relevant systems — Digital.ai Release for pipeline data, Jira for issue details, or your directory for team information.

3

Respond Results are presented as clear text or interactive cards, with source attribution showing where the data came from.

4

Remember Reva maintains three memory layers: semantic memory for facts and preferences across sessions, episodic memory that tracks past interactions with topics and entities, and procedural memory for behavioral rules. Follow-up questions work naturally — “Show my releases” then “Details for #3”. A relevance filter ensures only meaningful exchanges are memorized.

Integrations

Reva connects to your tools through the Model Context Protocol (MCP) — an open standard for AI–tool integration. Each integration is read-only by default and runs in an isolated container.

Digital.ai Release

Full visibility into your release pipelines: releases, phases, tasks, gates, teams, activity logs, and folder structures. Reva resolves your team memberships and shows you a personal dashboard of the releases you’re involved in.

Jira Cloud

Search and view issues, sprints, boards, and projects. Reva identifies you by your Jira account and can show “my tickets” without manual filtering. Cross-system linking finds Jira issues referenced in your releases.

Extensible

New integrations can be added by connecting additional MCP servers — no changes to Reva’s core are needed. Planned integrations include CI/CD systems (Jenkins, GitLab), ServiceNow, and Confluence.

Read-only by design: Reva can only read data from your systems. All write and delete operations are blocked at the integration layer.

Proactive Notifications

Reva doesn’t just answer questions — it proactively keeps you informed:

Release status changes — Get notified when a release starts, completes, or fails
Jira issue updates — Track important issue transitions in your Teams chat
Blocker alerts — Team members can send targeted alerts about blockers or urgent issues
Autonomous monitoring — Optional background checks for overdue tasks, failing gates, and stale approvals

Notifications are rate-limited and deduplicated to prevent alert fatigue.

Security

On-premises AI — The language model runs entirely on your infrastructure. No queries, conversations, or data leave your network for AI processing.
Azure AD authentication — Every message from Teams is cryptographically verified. Unauthorized requests never reach the application.
User authorization — Optionally map Teams users to your Release user directory for identity-aware responses.
Encrypted connections — TLS encryption for all external connections and database traffic.
Secrets management — All credentials stored as Docker or Kubernetes secrets, never in configuration files.
Read-only integrations — Reva cannot modify, create, or delete data in connected systems.

All connections to external systems originate from your infrastructure. No external system can initiate connections to Reva except Microsoft Teams (via Azure Bot Framework) and optionally your webhook sources.

Data Protection

Reva is designed for compliance with the EU General Data Protection Regulation (GDPR) and the German Works Constitution Act (BetrVG).

Privacy by design: All data processing occurs within your infrastructure. No data is transmitted to external AI services, analytics platforms, or the software provider.

No individual performance monitoring — Reva will never track, compare, or report on individual user activity. This is enforced at the AI level.
Anonymized activity logs — Summaries attribute actions to “a team member” rather than named individuals.
Works council ready — Built-in privacy guardrails are designed to facilitate BetrVG §87 approval.

Deployment

Reva runs on your infrastructure as a containerized application. Two deployment models are supported:

Docker Compose — Single-server deployment for small teams or evaluation. All services (Reva, database, cache, LLM) run on one machine.
Kubernetes — Production deployment with TLS ingress, network policies, and horizontal scaling. Integration servers run as sidecars for isolation.

Hardware requirements: A single server with a 16 GB GPU (e.g. NVIDIA RTX 4060 Ti or better) can serve a team of up to 50 users. The AI model runs locally — no cloud GPU rental needed.