Instances & architecture
Every ExoHuman Cloud customer gets a dedicated instance. There is no shared, multi-tenant application where several customers’ data sits side by side in one database: your users, agents, conversations, files, credentials and audit trail live in an instance that belongs to you alone.
What an instance contains
Section titled “What an instance contains”| Component | Role |
|---|---|
| Web app | The ExoHuman interface: Workspace, Dashboard, agent pages, Operations and Settings. |
| Backend API | The REST API used by the web app, the CLI and integrations. |
| Scheduler | Fires cron schedules, loops and reminders exactly once. |
| MCP server | Exposes the platform as MCP tools at https://<subdomain>.exohuman.cc/mcp. |
| PostgreSQL | Your instance’s database. |
| Redis | Queues, locks and live event streams. |
| Agent containers | One container per agent, built from the release’s agent base image. |
All of these run one pinned ExoHuman release. Your instance moves to a new release only during a planned update (see Uptime & SLA).
Control plane and worker pools
Section titled “Control plane and worker pools”An instance is split into two layers:
- The control plane — web app, API, scheduler, MCP server, database and cache. It holds state and decides what runs where.
- Agent execution — agent containers, scheduled across elastic worker pools. Each pool provides one kind of capacity (for example Standard CPU, High memory or a GPU type) and grows and shrinks with demand. See Autoscaling & worker pools.
On Launch and Scale, your control plane runs on our managed infrastructure in your chosen region, isolated from other customers. On Enterprise, the control plane is dedicated hardware in the region you choose, with optional private networking (VPC peering, PrivateLink or VPN).
Isolation
Section titled “Isolation”- Per-customer instance. Separate database, cache, secrets and storage per instance.
- Per-agent containers. Each agent has its own filesystem, process space, resource limits and credentials. Agents cannot read each other’s files.
- Agent-to-agent communication goes through the platform and only along permissions you grant.
- Network. Agent traffic to the internet leaves through the instance’s egress; inbound access to agents is only through the platform.
- Credentials are encrypted at rest and injected into the agent that needs them. See Credential management.
Lifecycle of an agent container
Section titled “Lifecycle of an agent container”- Created when you create the agent, with the compute profile you picked.
- Started when it has work: a chat message, a schedule, a delegation, a webhook or a channel message.
- Suspended when idle. Its workspace and memory are preserved; compute billing stops.
- Resumed transparently the next time it has work.
Agent memory and workspace history are versioned in git, so suspending or moving a container never loses work. See Memory & git state.
Your URL and domains
Section titled “Your URL and domains”Your instance is served at https://<subdomain>.exohuman.cc with a TLS certificate we manage. The URL is shown in your account dashboard. Custom domains are available on Enterprise.
What we operate for you
Section titled “What we operate for you”- Provisioning, capacity and autoscaling.
- Operating-system and container-runtime patching.
- ExoHuman updates in announced windows, with pre-update backups and automatic rollback.
- Daily backups and restore tests. See Backups & recovery.
- Monitoring and incident response. See Uptime & SLA.
You remain the administrator of what happens inside your instance: people, roles, agents, integrations and policies.