AgenticMail started as a way to give AI agents email. AgenticMail Enterprise is what happens when you take that idea and ask “what if agents had everything a human employee has?”
Email, phone, calendar, browser, memory, file storage, team communication. If a knowledge worker uses it, an agent should have access to it too. That’s the core thesis behind the Enterprise platform, and building it meant rethinking the entire architecture from the ground up.
The 6 Tier Architecture
Enterprise is organized into six layers, each handling a distinct responsibility:
Tier 1: Infrastructure. Database backends, message queues, file storage, secrets management. This layer is swappable. We support 10 different database backends because enterprise customers have strong opinions about their data layer, and they should.
Tier 2: Identity and Auth. Agent identities, API keys, role based access control, tenant isolation. Every agent is a first class identity in the system with its own credentials, permissions, and audit trail.
Tier 3: Communication. Email (IMAP/SMTP), phone (via telephony integrations), SMS, calendar, and team messaging. This is where the original AgenticMail lives, expanded with additional channels.
Tier 4: Capabilities. Browser automation, file manipulation, code execution, web search. These are the “hands” that let agents interact with the digital world beyond communication.
Tier 5: Intelligence. LLM integration, memory systems, context management, tool orchestration. This layer handles the reasoning engine and everything the agent needs to think and remember.
Tier 6: Management. Dashboards, monitoring, billing, lifecycle management, fleet operations. The control plane for running agents at scale.
Each tier can be updated, scaled, or replaced independently. The interfaces between tiers are stable contracts, so upgrading the database backend doesn’t require changes to the communication layer.
82 Engine Modules, 270+ Tools
The platform ships with 82 distinct engine modules that compose together to form agent capabilities. Each module is a focused unit: one handles IMAP connections, another handles calendar sync, another manages browser sessions.
These modules expose over 270 tools that agents can use. A tool might be “send email,” “schedule meeting,” “search the web,” “read a PDF,” or “remember this fact for later.” The tool surface is large because a digital employee’s job is large. You build a comprehensive toolkit and let the agent’s reasoning engine decide what to reach for.
Not every agent loads every tool. That would be wasteful and confusing (more on this in a future post about smart tool loading). But every tool is available in the platform for any agent that needs it.
10 Database Backends
Supporting 10 database backends sounds excessive until you talk to enterprise customers. One company standardized on CockroachDB. Another uses Aurora Postgres. A startup wants embedded SQLite so they can run the whole thing on a single machine.
The database abstraction layer normalizes all of this behind a common interface. Agents and engine modules don’t know or care which backend is active. They call the same functions, and the abstraction layer translates to the appropriate dialect.
This was one of the more painful engineering efforts in the project. SQL dialects differ in subtle ways (date functions, upsert syntax, JSON handling), and getting all 10 backends to pass the same test suite required careful conditional logic. But the result is genuine deployment flexibility.
Single npx Command Setup
Despite all this complexity, getting started is one command:
npx agenticmail_enterprise initThis scaffolds the configuration, sets up the default SQLite backend, provisions an admin account, creates a first agent, and starts the server. You can go from zero to a running agent with email, calendar, and browser access in under two minutes.
Digital Employees, Not Chatbots
The framing matters. AgenticMail Enterprise treats agents as digital employees, not chatbots. They have identities, schedules, communication channels, memory, and accountability via audit logs. They can be promoted (given more capabilities), put on leave (suspended), or terminated (deprovisioned).
This isn’t just branding. The workforce metaphor drives real architectural decisions. Employees need onboarding (agent provisioning). They need performance reviews (monitoring and evaluation). They need to coordinate with teammates (multi agent communication). Building the platform around this metaphor keeps the feature set grounded in what actually matters for production deployments.
Source Code
The AgentState type captures the full lifecycle of a digital employee, from initial draft through active operation to eventual teardown:
export type AgentState =
| 'draft' | 'configuring' | 'ready' | 'provisioning'
| 'deploying' | 'starting' | 'running' | 'degraded'
| 'stopped' | 'error' | 'updating' | 'destroying';