Building a Local-First Agent Platform Where the Vendor Is Never in the Middle

Description and use

TheYgent delivers a local-first agent platform that removes vendor dependency, simplifies model management, and ensures data privacy.

Problem

Visual AI builders require users to manage model engines, quantization, and server infrastructure, leading to complexity and vendor dependency.

Outcome

TheYgent delivers a fully local, privacy‑preserving agent platform that removes vendor dependency, simplifies model management, and supports durable, observable workflows.

Workflow

• 1. Deploy inference plane on local hardware (laptop, GPU box, or both).
• 2. Deploy control plane (FastAPI app) on the same host or cluster.
• 3. Register a model via the inference plane’s admin API, binding a logical ID to an engine, source, and modality.
• 4. Build agents using the editor; agents reference logical IDs, not engine names.
• 5. Run agents; control plane sends model calls to inference plane via shared HTTP client.
• 6. Manage engine lifecycle: lazy start, resident ceiling, idle reaping, and cancellation of abandoned calls.
• 7. Use durable runtime for unattended tasks, leveraging DBOS and Postgres for state journaling.
• 8. Monitor runs via Postgres spans and optional OpenTelemetry export.
• 9. Verify privacy by reviewing the complete list of network connections made by the platform.

Evidence

Every visual AI builder we tried had the same footnote: local models supported — bring your own server.

Source

The inference plane runs models. It holds everything heavy — engine processes, model weights, the model registry, API credentials — and it runs wherever you point it: the same laptop, a GPU box across the network, or both.

Source

The control plane never imports an engine. There is no MLX, vLLM, or llama.cpp binding anywhere in it.

Source

Binding an ID to something that serves it happens once, on the inference plane's management API...

Source

The inference plane doesn't ask you to start servers. It starts them.

Source

The durable runtime provides that. It is built on DBOS, an embedded Python library that journals workflow state into the Postgres the control plane already uses — no new stateful service to deploy.

Source

Every run produces spans and per-node inputs and outputs in your own Postgres, and the interface draws them as a zoomable waterfall...

Source

The list includes the one request we could have left out: at startup, the inference plane's model client library fetches a public model price list from GitHub.

Source