ToolBB-0052

Edge Agents: an offline agent runtime for Linux edge boards

Edge Agents, from ForestHub, runs AI agent workflows offline on Linux edge boards in a container of about 30 MB. Workflows mix GPIO, ADC, PWM, UART and MQTT nodes with local small language models or cloud LLMs. The README says 5 boards, including a Raspberry Pi 5 and a Jetson Orin Nano, were exercised on the authors' bench.

Open original source github.com · ForestHub AI

Claimed
Why this evidence label?

README target table marks several boards as exercised on the authors' bench; the hero GIF shows the editor, not the hardware.

How we check entries
Preview image for Edge Agents: an offline agent runtime for Linux edge boards
The source's own preview image, from github.com.

Briefing

fh-workflow open starts a visual editor for building workflows, and fh-workflow deploy packs one into a docker-compose bundle for offline install. GPIO, ADC, DAC and PWM nodes go through Linux character devices. Local models run through llama.cpp or any OpenAI-compatible endpoint; cloud nodes cover Anthropic, OpenAI, Gemini and Mistral. The README compares the Go engine's image with n8n, Dify and OpenClaw images of 500 MB to 1 GB.

The other boards on the bench list are an x86 NUC, an STM32MP25 and a Bosch Rexroth ctrlX CORE. The README says there is no per-device CI. The engine is AGPL-3.0 or commercial; the contract and workflow-core packages are Apache-2.0.

From the GitHub API

Repository facts

Repository facts for foresthubai/edge-agents, from the GitHub API
Repositoryforesthubai/edge-agents
Stars106
LicenseAGPL-3.0 Copyleft
LanguageTypeScript
Last push24 Aug 2026 2 mo ago

Tags

Interface

Report a problem with this entry