- Modules
- One core holds every source, fact and permission. Each team adds a pack: the tools it reads, the results it gets, what starts them and the actions it may propose. A pack can run on its own stack. Packs share the core’s memory and permissions, so you add, swap or remove one without touching the rest.
- Connectors
- Slack, Microsoft Teams, Google Workspace, Microsoft 365, Notion, Confluence, GitHub, Linear, Jira, HubSpot, Salesforce, Zendesk, Stripe and Zoom, through OAuth.
- Formats
- Messages, documents, PDFs, slides, spreadsheets, screenshots, code, and meeting recordings and transcripts, in one index.
- Retrieval
- Keyword and vector search over passages, plus a walk of the graph of people, projects, customers and decisions. A reranker keeps only the strongest evidence.
- Memory
- Every fact keeps when it was true and where it came from, so Hatchling can say what changed and which source is newer.
- Permissions
- Access lists sync from every source and apply before retrieval. Text you can’t open never reaches the model.
- Reasoning
- Hard requests are worked in steps: plan, search several tools, draft, check every line against its sources, cite.
- Actions
- Updates the doc, files the ticket or sends the follow-up in the tool where the work lives. You approve each kind of action, and every change is logged with its sources.
- Delivery
- The desktop app, Slack, email, the web app, a REST API and webhooks.
- Desktop app
- For Windows, macOS and Linux. A notification when a result is ready, and a shortcut to ask from anywhere. The web app is the same, in a browser.
- Agents
- MCP for context and tools, A2A to hand over a whole task, and the open Agent Skills format for playbooks. Every request carries the permissions of the person it works for.
- Hosting
- In the EU: the app, the database and every model call. One company’s data never reaches another.