Introduction

TaTi is an open source platform built for delivery, SRE, and ops teams that want an AI copilot wired to their real toolchain — not just generic text.

The problem

Without context, a generic assistant cannot:

• read your state in Slack or Discord;
• run controlled SQL on PostgreSQL;
• read your OpenMetadata catalog or Grafana dashboards;
• open a merge request on GitHub / GitLab.

TaTi is an orchestration layer: a single UI where you configure MCP servers (Model Context Protocol). Each bridge exposes tools the model can call, with URLs and secrets you control.

Main components

Item	Role
Web app	Chat, user settings, MCP list, optional local auth.
PostgreSQL	App persistence (sessions, MCP server config in the product).
MCP services	One process (or container) per tool family: see MCP connectors.

It is not a closed “app store”: everything in the repo’s docker-compose.yml can be enabled or not depending on your .env and token availability.

UI overview

Sign-in screen

When local authentication is enabled (TATI_AUTH_REQUIRED), the first page shows the Sign in form (email, password), the TaTi logo, and a side area for product visuals or messaging.

Home after sign-in

After login, home includes the chat thread, the sidebar (new chat, thread list, profile, active MCP servers, settings), and a central product area (version, quick access to a conversation and settings).

Settings — AI providers

Settings lets you configure model providers (Claude, OpenAI, Mistral, Ollama, etc.): API keys, default model, temperature, max tool iterations, and the default provider (“Default” badge). Other tabs cover MCP servers, account, users, and an embedded quick start.

Images are indicative; theme, labels, and tab layout may change with versions.

New MCP server — presets

When you add a server from Settings → MCP servers, New MCP server shows a grid of presets: ready-made integrations (databases, clouds, messaging, observability, etc.). Fine-tuning (URL, secrets, Compose ports) is in the MCP connectors reference.

Image indicative; the preset list may evolve with versions.

Suggested reading order

1. Quick start — run Postgres + app + a few MCPs locally.
2. Architecture — understand flows (browser → app → MCP).
3. Configuration — stabilize .env (no duplicate keys, secrets out of Git).
4. MCP reference — configure each connector (ports, /mcp URL, OAuth headers).

Glossary

• MCP: open protocol to expose tools / resources to an AI client; TaTi acts as a client (via its backend) to your bridges.
• Streamable HTTP: HTTP transport used by repo bridges for /mcp.
• DATABASE_URL: Postgres connection for the TaTi app; distinct from the Postgres MCP connection used for conversational SQL.

---

Next: Quick start for concrete commands.