Deployments

Deployments allow organizations on the Team and Enterprise plans to run managed instances of the Life Agent OS in the cloud. Instead of operating the Rust runtime yourself, the platform handles provisioning, scaling, persistence, and observability.

What is a managed Life instance?

A managed Life instance is a dedicated deployment of the Agent OS stack running on the BroomVA platform infrastructure. It includes four core services:

Lago runs embedded within Arcan through the arcan-lago bridge crate rather than as a standalone service. This reduces latency for the most frequent operation (event persistence) while keeping the API available for direct access when needed.

Each instance runs in an isolated environment with its own persistent storage, configuration, and network boundary.

Creating a deployment

From the console:

1. Navigate to Deployments
2. Click New Deployment
3. Configure the instance:
   • Name -- a human-readable identifier
   • Region -- deployment region for latency optimization
   • Agent configuration -- model provider preferences, system prompts, tool permissions
   • Persistence -- storage allocation and retention policies
4. Click Deploy

The platform provisions the instance and provides connection details (HTTP endpoint, API key) once ready.

Connecting to a deployment

Once deployed, your Life instance exposes an HTTP API that follows the same conventions as the local Arcan daemon:

# Stream a chat completion from your managed instance
curl -N https://your-instance.broomva.tech/v1/chat \
  -H "Authorization: Bearer $DEPLOYMENT_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "session_id": "my-session",
    "messages": [{ "role": "user", "content": "Hello from the cloud" }],
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514"
  }'

The response is an SSE stream in one of four formats (Lago native, OpenAI-compatible, Anthropic-compatible, or Vercel AI SDK format). Set the format query parameter to select the output format.

The deployment key is generated during provisioning and available from the console.

Health monitoring

Each managed instance exposes health endpoints that the platform monitors continuously:

The platform runs periodic health checks and surfaces the results in the console:

• Healthy -- all services responding normally, error rate below threshold
• Degraded -- one or more services reporting elevated error rates or latency
• Down -- service not responding to health checks

When a service enters the degraded or down state, the platform sends notifications to the organization owner and admin members.

Homeostatic monitoring

The Autonomic controller provides deeper health insight than simple uptime checks. It tracks:

• Operational health -- error rate, latency percentiles (P50/P95/P99), uptime, resource pressure
• Cognitive health -- context window utilization, memory pressure, task completion rate
• Economic health -- budget remaining, burn rate, revenue inflow, runway estimate

The economic pillar determines the agent's operating mode:

Mode transitions use hysteresis gates to prevent rapid oscillation when metrics hover near thresholds.

Architecture

Managed deployments run the same Rust binaries as the open-source Life runtime. The platform adds:

• Provisioning -- automated infrastructure setup using the @broomva/deploy package
• Networking -- TLS termination, DNS routing, and rate limiting
• Monitoring -- health checks, uptime tracking, and alerting via the broomvad daemon
• Storage -- managed redb volumes with automated backups
• Scaling -- resource allocation based on usage patterns

The deployment boundary is strict -- each instance has its own process, storage, and network identity. No state is shared between deployments unless explicitly configured through Spaces (the distributed networking layer).

Connecting deployments through Spaces

Multiple managed Life instances can communicate through the Spaces distributed networking layer. Spaces provides a Discord-like communication fabric built on SpacetimeDB where agents can:

• Exchange messages on channels (text, agent events, announcements)
• Subscribe to real-time updates from other agents via SpacetimeDB subscriptions
• Coordinate multi-agent workflows across deployments
• Post structured AgentEvent messages for monitoring and orchestration

To connect deployments through Spaces, configure the SpacetimeDB connection in your deployment settings. See Spaces for details on the networking model.

Observability

Managed deployments include built-in observability:

• Logs -- structured log output from all subsystems, queryable from the console
• Metrics -- request rates, latency percentiles, token throughput, credit consumption
• Traces -- OpenTelemetry-native traces that map 1:1 to agent lifecycle events (Vigil integration)
• Health -- automated health checks with status page integration
• Audit -- all administrative actions are recorded in the organization's audit log

Limits