Product updates

The ForestHub newsletter is live with a proper double opt-in flow. Signup lives in the footer of every page, followed by a confirmation email and a dedicated confirmation page. Product updates and edge AI notes arrive on a steady rhythm.

ForestHub is now positioned as the orchestration layer for edge agents. Workflows run as a deterministic graph on a Linux gateway, with deterministic operations and LLM agents wired into the same canvas. The graph is the program and the model is one node among many.

A dedicated page now answers a question that kept coming up in conversations. It walks through what an edge agent is, why it runs on the device instead of the cloud, and where ForestHub fits. Available in English and German.

The platform page now states the open core model plainly and links to the public edge agent runtime on GitHub. The code is there to read before anyone signs up.

The Visual Builder now stands on its own as a self-contained package. Undo, redo, save, and unsaved change tracking live inside the builder itself, which keeps the canvas predictable and quick to work in.

The engine was decoupled from the backend and split into standalone packages, with the engine and the LLM proxy each able to run on their own. A shared contract layer keeps the Go and TypeScript sides generated from the same definitions, so the pieces stay in step. The platform can now run locally without an account.

A command line companion now opens, validates, and updates workflow files from the terminal. It runs the local builder against a workflow file, which keeps quick edits and checks close to where the work already happens.

The edge agent runtime is now public on GitHub under AGPL-3.0. It is a compact runtime that runs offline by default, with GPIO, UART, and MQTT as first class nodes. Available to read, run, or build on.

A new article looks at embedded AI as a virtual service technician, written with our partner Grossenbacher Systeme. It covers how small and large language models combine with agentic orchestration in the field.

A new section on the home and platform pages lays out the intelligence cascade, a five step model for deciding how much intelligence a task actually needs. It runs from simple rules up to full language models, so the heavy option is the last resort rather than the default.

Three pages went live together to make the graph first idea concrete: a patterns page, a roadmap, and an engine page. Together they show what ships today, where it is heading, and how a workflow graph turns into something that runs.