A local-first, agent-assisted workspace where migration engineers author pipelines, operate deterministic workflows, monitor live execution, and collaborate with the model of their choice.
Ask questions, inspect the active migration, draft changes, load project memory and skills, and move through the work conversationally. Read operations flow naturally; writes stop for explicit human approval.
The Operator workspace talks directly to the local engine over authenticated gRPC. Run states, preview SQL work, and inspect history through auditable commands that do not depend on an AI provider.
Desktop brings the full conversion lifecycle into one application while keeping each workspace clear about what it reads, what it can change, and which engine contract it uses.
Chat with a migration-aware agent, resume persisted conversations, inspect live pipeline context, and review approvals inline.
Run states, dry-run actions, emit work, and inspect recent runs through direct, audited engine commands—no model required.
Follow live throughput by pipeline state, inspect run history, and drill into the same engine data available to Copilot.
Author conversion processes on a visual canvas with connectors, states, actions, an inspector, and MetaMap™ editing.
Configure connectors, materialization, schedules, checkpoints, source tables, and supporting migration resources.
Review audit activity, manage local users and roles, configure providers, and administer offline-friendly licensing.
Swapping models never changes the migration engine or its tool contract. Every call is authenticated, role-checked, and routed through the same deterministic substrate.
Reviews intent, approves writes, and retains final authority.
Reasons over context and proposes MCP-native actions.
Authenticates, authorizes, schedules, executes, and audits.
Owner-only engine tokens and local app sessions authenticate every non-health call.
The engine re-checks the caller’s role before every tool dispatch.
Per-connection rate and concurrency controls keep one session from starving another.
The Electron application manages a bundled .NET 8 engine that owns orchestration, executors, connectors, scheduling, security, licensing, and the dual-dialect control plane.
The user interface can change. The engine remains the deterministic source of truth for execution.
Desktop separates model transport from authentication. Choose a hosted API, use an eligible subscription sign-in, or keep inference local through Ollama or LM Studio. The approval gate and MCP tool surface do not move.
Native Anthropic, OpenAI-compatible providers, and supported subscription flows.
Keyless loopback operation through Ollama or LM Studio keeps model traffic on the machine.
Model choice changes the transport—not the engine, tools, permissions, or migration state.
Desktop turns proven project knowledge into governed local assets that can be reviewed, bound to the right context, and reused across migration work.
Persist project facts and decisions locally, with proposals reviewed before they become durable context.
Create reusable migration guidance and bind it to pipelines, connectors, source systems, or MetaMap™ assets.
Preview and import reusable pipeline configuration with connector remapping and approval before writes.
Move mapping definitions through a preview-first package flow while keeping credentials outside the artifact.
RSA-signed local licenses support disconnected environments. Account connection and entitlement sync can be added when useful; hosted connectivity is not the runtime foundation of the application.
Tell us about your source systems, operating boundaries, and migration team. We’ll help determine whether AetherFlow Desktop fits the environment.