Bare Metal and Virtualized Deployment
Run mBedLM as a local-first runtime on commodity hardware, virtualized nodes, or staged enterprise environments without changing core orchestration behavior.
Core Features
Deterministic routing
mBedLM selects execution paths using policy-aware routing. Primary local routes and fallback provider routes are configured explicitly so runtime behavior remains predictable under load and failure conditions.
- Keep the local provider first when the task can be served on-device.
- Use explicit provider health checks before activating fallback routes.
- Treat route selection as a policy decision, not an implicit retry loop.
Stable response contracts
Response normalization shields downstream services from provider variance. This allows teams to evolve provider choices without breaking API consumers or post-processing steps.
Route -> Normalize -> Validate -> Handoff
Runtime guardrails
Startup checks validate environment and route readiness before enabling orchestration. Trust boundaries and rollout gates separate exploratory behavior from production-safe flows.
graph TD A[Validate config] --> B[Check provider health] B --> C[Confirm route policy] C --> D[Start runtime] D --> E[Admit workloads]
System Architecture
The deployment model follows a strict lifecycle: validate config and secrets, initialize serving/runtime layers, activate route policy, then admit workloads. This sequence minimizes hidden startup failures and route drift.
graph TD A[Secrets and config] --> B[Serving layer] B --> C[Route policy] C --> D[Workflow admission] D --> E[Observability and rollback]
Deployment Models
- Single-node local development for fast iteration and model testing.
- Virtualized team staging for route, contract, and fallback validation.
- Phased enterprise rollout with governance checks and support workflows.
# Example deployment gate sequence mbedlm providers list mbedlm route test --task analysis --policy strict mbedlm workflow run --profile enterprise-rollout docker compose up -d