What is amahub
amahub is a managed workspace where a company’s people delegate real work to AI agents — inside the company’s own tools, under each person’s own access, with the controls an organization actually needs.
An agent session in amahub runs in an isolated cloud sandbox with no direct internet: everything it touches goes through the org’s connectors or the egress proxy, both of which enforce policy and write an audit trail.
The shape of the product
Section titled “The shape of the product”| Surface | What it is |
|---|---|
| Tasks | The chat-first home: type what you would hand a colleague; the agent works in the background and survives your laptop closing. |
| Processes | A task your team runs repeatedly, saved with the profile it must run with — model, connector allowlist, skills, standing instructions. |
| Automations | Processes (or free instructions) on a trigger: a schedule or an event. Runs are ordinary sessions, visible to everyone. |
| Connectors | Company tools the agent may use — an official catalog (OAuth or tokens) plus custom MCP servers with vaulted org credentials. |
| Skills | Reusable capability packs shared to the org, a group, or a person. |
| Security | Live egress monitor, an alert queue, deterministic policies (built-in and custom, with DLP content detectors), and outbound alert routing. |
| Audit | Every session event, queryable and exportable — including by external tools via temporary tokens. |
Design principles
Section titled “Design principles”- Least privilege by construction — a session can only reach connectors granted to the person it runs as; a process can pin an even narrower allowlist.
- Everything lands in the audit log — including reads of the audit log.
- Multiplayer by default — sessions can be shared, good sessions become processes, processes become automations the whole team sees.
- Familiar first screen — the home page is a chat composer, not a dashboard.