Abacus.AI External Service Connectors
Abacus.AI connects to your data sources and business applications through connectors. Every connector belongs to one of two families, and the first decision you make is which family fits your use case:
| User Connectors | Organization Connectors | |
|---|---|---|
| Who sets it up | Each end user, for themselves | An admin or developer, once for the whole organization |
| Whose identity is used | The signed-in user's own account | Shared credentials, a service account or a service principal |
| What the data is used for | Answering questions live in ChatLLM Teams chats (RouteLLM), the Abacus AI Agent and Custom Chatbots | Ingesting data into datasets and feature groups for ML models, GenAI document sets and Custom Chatbots |
| When data is read | In real time, on every request | On a schedule when the dataset refreshes, or continuously with Realtime Synced Datasets |
| Access control | The source system's own permissions apply automatically, per user | Everyone with access to the trained model sees the same data, unless you use a permission-aware (RBAC) connector |
| Where to configure | ChatLLM Teams → Profile → User Connectors | Abacus.AI → Connected Services Dashboard |
User Connectors
A user connector links your own account on a service (Gmail, Slack, SharePoint, Snowflake and so on) to Abacus.AI. Everything the connector does runs under your identity, so you can only ever see what you already have access to, and the source system records the activity against your account.
Use a user connector when:
- Each person should see only their own data, for example their own mailbox, their own tickets or the database rows their role allows.
- You want a live answer from the source system rather than a periodically refreshed copy.
- You want to ask questions in ChatLLM Teams right away, without an admin building a project first.
Most user connectors work immediately in chat with RouteLLM and the Abacus AI Agent. Some need an admin to create an organization-level config connector first, which stores the OAuth application or scopes that every user connection then inherits. Each connector page explains whether that applies.
Organization Connectors
An organization connector is created once by an administrator or developer and is shared by the organization. It ingests data into Abacus.AI so it can be used to train machine learning models, build document sets for GenAI use cases, and power Custom Chatbots.
Use an organization connector when:
- You are building a machine learning model or a retrieval-augmented (RAG) chatbot over a defined set of documents or tables.
- The data should be refreshed on a schedule, or synced continuously, rather than fetched on every question.
- A developer needs full control over what is ingested, how it is transformed, and which version is used for training.
Organization connectors are grouped by the kind of source they read: Database, File, Application and Messaging connectors, plus Permission Aware Connectors (which keep the source system's access rights on ingested documents) and Realtime Synced Datasets.
Organization connectors are by default available only to developers who have access to the development environment of the platform. You can further restrict them with the RBAC Module.
→ Organization Connectors guide
Which one do I need?
| I want to… | Use |
|---|---|
| Ask ChatLLM about my own emails, files, tickets or meetings | User connector |
| Let a Custom Chatbot run SQL under each user's own database role | User connector (database) |
| Train a forecasting or classification model on a database table | Organization connector |
| Build a chatbot over a fixed set of SharePoint or Confluence documents | Organization connector |
| Build that same chatbot, but keep the source system's per-document permissions | Organization connector with permission awareness (RBAC) |
| Keep a dataset continuously in sync with Jira, Confluence, SharePoint or Teams | Organization connector with Realtime Synced Datasets |
Microsoft connectors need tenant consent
For Microsoft connectors (OneDrive, SharePoint, Outlook, Microsoft Teams, Azure Storage, Azure SQL), at both the user and the organization level, a Microsoft Entra admin must create the connector first and tick "Consent on behalf of your organization" on Microsoft's consent screen. This is not the same as the Grant admin consent button in the Entra admin center. See Microsoft Entra Tenant Consent for details.
MCP Servers
In addition to connectors, you can connect Abacus.AI to external tools and data through Model Context Protocol (MCP) servers. MCP servers can be added by an individual member (user-level) or configured once by an admin and shared across the whole organization (organization-level), and they can be attached to Custom Chatbots. See MCP Servers for details.
Network Allowlisting
If your data source sits behind a firewall, you may need to allowlist the Abacus.AI egress IP addresses so our connectors can reach it. See Egress IPs for the addresses to allow, for both Enterprise and Self-Serve plans.