Agentwork is a shared knowledge layer for companies. It connects to the tools where work already lives, indexes their content, and lets anyone on the team ask a question in plain language and get an answer that links back to the source documents.
When the answer is not in any connected source, Agentwork identifies the person most likely to know, sends them the question, and stores the reply so the next person who asks gets it directly. When two sources contradict each other, it flags the conflict and routes it to the relevant owner instead of picking one.
Access follows the permissions of the underlying tools: a user only sees answers drawn from sources they can already open, and shared and private knowledge are kept in separate layers with the boundary enforced in code rather than by the model. Typical queries include resolving policy questions, finding an owner, summarizing a review meeting, tracking a project across tools, and preparing a customer briefing.
Features
- Connectors: Slack, Notion, Google Workspace (Docs, Slides, Sheets, Gmail), and Fireflies today, with GitHub, Linear, Microsoft Teams, and Raycast in development
- Sourced answers: every response cites the documents it drew from so you can verify it
- Gap filling: unanswered questions are routed to the likely expert and their reply is saved for reuse
- Conflict handling: contradictory sources are surfaced and reconciled with the owner rather than guessed
- Permission-aware retrieval: answers respect the source tools' existing access controls
- Workflows: scheduled recurring briefings and role-specific repeatable tasks run over the same knowledge base
- MCP interface: a Model Context Protocol endpoint exposes the knowledge layer to coding agents and other MCP clients
- Hosting: cloud-based, with encryption in transit
