mBedLM 2.0
Model-agnostic orchestration layer with multi-agent patterns, retrieval-aware routing, and local-first deployment compatibility across Q4, Q5, and Q8 quantization levels.
Four tightly integrated layers that make local AI reliable at any scale.
Product map
mBedLM, the CLI, the SDK, and VIE are designed as separate surfaces with one shared orchestration spine. That keeps the product coherent while letting operators choose the surface that matches their workflow.
graph TD A[mBedLM CLI] --> B[VIE orchestrator] A --> C[mBedLM SDK] B --> D[Routing policy] B --> E[Memory and adapters] C --> D C --> E D --> F[Stable response contracts]
deployed_code
route
memory
model_training
code_blocks
terminal
data_object
dns
commit
device_hub
dashboard_customize
settings_input_component
mBedLM CLI
Operational command surface for local teams: environment setup, runtime controls, provider configuration, and rapid workflow invocation for agentic tasks.
mBedLM SDK
Developer toolkit for integrating orchestration, memory controls, and agent scheduling into custom applications. Type-safe interfaces for Python and TypeScript.
VIE
Primary orchestrator experience for complex request execution, model fallback policy, and controlled escalation from fast-path to deep multi-agent processing.