The connector catalogue is the boring part of the Atlas value story. It is also the part that decides whether the agent you ship next quarter is a week of configuration or a quarter of integration engineering.
What is in the catalogue today, what each connector reads and writes, and what it costs (in operator time) to enable.
Productivity and collaboration
- Microsoft Graph (Microsoft 365 + Teams + SharePoint + Outlook). Read calendar, mail, Teams channels, SharePoint sites, OneDrive. Write Teams messages, calendar events, Outlook drafts.
- Google Workspace (Gmail, Drive, Calendar, Docs, Sheets). Read mail, docs, calendar. Write Gmail drafts, Calendar events.
- Slack. Read channel history, threads, DMs (scoped by permission). Write to channels, threads, ephemeral messages.
- Discord. Read channel history, threads. Write to channels with threading support.
- Notion. Read pages, databases. Write to pages and database rows.
- Confluence. Read spaces, pages, attachments.
CRM and sales
- Salesforce. Read accounts, contacts, opportunities, custom objects. Write activity records, notes, opportunity updates, contact updates.
- HubSpot. Read contacts, deals, companies, tickets. Write contact properties, deal updates, notes.
- Zoho CRM. Read modules including leads, contacts, deals, accounts. Write records and notes.
- Pipedrive. Read deals, organizations, persons. Write notes and activity records.
Banking, finance, asset management
- BlackRock Aladdin. Read portfolio data, positions, performance metrics. Read-only by design; orchestration of writes through approved workflows only.
- Avaloq. Read client positions, orders, mandate data. Write through approved channels with full audit-log capture.
- Generic banking core (read-side). SWIFT message reading, transaction history retrieval through SFTP / MFT patterns common in regulated environments.
Operations, supply chain, ERP
- SAP S/4HANA (read-side connector). Read materials, orders, supplier records, inventory.
- NetSuite. Read records and saved searches. Write to record types per the customer’s permission model.
- Generic shipping API pattern. Schedule retrieval, capacity reads, container booking write-side through customer-specific implementations.
Developer-facing and community
- GitHub. Read issues, pull requests, discussions, releases. Write comments, issue updates, PR reviews.
- GitLab. Read issues, merge requests. Write comments and updates.
- Linear. Read issues, projects. Write issue comments, status updates.
- PostHog / analytics events. Read event streams scoped to the customer’s analytics workspace.
Support and ticketing
- Zendesk. Read tickets, conversations, knowledge base. Write ticket comments, status updates.
- Intercom. Read conversations, users, articles. Write conversation replies (with approval) and notes.
- Jira Service Management. Read tickets, comments, SLA data. Write comments and status changes.
Knowledge and content
- Docling (document parsing). Parse PDFs, Word documents, PowerPoint, complex tables, scanned documents with OCR. Hands structured content to the substrate for indexing and retrieval.
- Web crawler. Crawl customer-approved domains for documentation, knowledge base content, pricing pages.
- RSS / Atom feeds. Read scheduled feed updates for change-detection workflows.
Generic patterns (the long tail)
- HTTP / REST connector framework. Wire any REST API into Atlas with declarative configuration. Includes authentication patterns for OAuth2, API key, JWT, bearer-token, basic auth.
- GraphQL connector framework. Same pattern, GraphQL query / mutation support.
- Generic SQL database connector. Postgres, MySQL, SQL Server, Snowflake, BigQuery. Read queries with parameterization; writes through approved templates only.
- SFTP / file-glob connector. For systems that drop files into a staging directory rather than expose an API.
- Webhook in / webhook out. Receive events from any system, fire actions to any endpoint.
What every connector provides by default
A new connector is not done until it provides four things, regardless of which system it talks to:
- Source-receipt mapping. Every piece of data the connector retrieves can be cited back to the source system, document, and version.
- Approval-gate integration. Write-side operations go through the substrate approval queue. The agent drafts; the operator approves; only then does the connector execute the action.
- Audit-log instrumentation. Every read and every write generates a structured audit entry tagged with the tenant, the agent, the action, and the result.
- Tenant-scoped credentials. Connector credentials live in the customer’s tenant; the substrate cannot accidentally use one tenant’s credentials to call into another tenant’s connector.
These are not features; they are substrate properties. A connector that lacks any of the four does not ship.
What is not in the catalogue (yet)
A few connector categories where the pilot demand has not been high enough to prioritize, but where the framework will produce a connector if the customer brings the requirement:
- Marketing automation (Marketo, Pardot)
- Specialized healthcare EMRs beyond the standard FHIR pattern
- Specialized legal practice management systems
- Long-tail proprietary banking cores beyond Avaloq and Aladdin
The framework is the same; the integration timeline depends on the protocol and the customer’s test environment access.
How customers think about the catalogue at procurement time
The honest version of the procurement question: “Which of the systems we already pay for will Atlas read from and write to on day one?” The answer almost always covers the customer’s top five systems out of the box. The exception is regulated industries with proprietary backends, where one or two connectors need to be built in the first pilot phase. The framework keeps that work in days, not quarters.
For the substrate that makes the catalogue compose well, read What Is Atlas?. For the operator playbook on scaling from one agent to many integrated agents on the catalogue, read Scaling From Your First to Your Fifth AI Agent. For the customer evidence of catalogue-driven scaling, read How Legacy Went 8x in 12 Months on Atlas.