All plugins

model-forge-image

Model Forge local image-generation backend — routes Hermes image_generate to any OpenAI-compatible images API (media gateway, vLLM-Omni, SGLang Diffusion, ComfyUI). Companion package for the Model Forge plugin. Disclosure — sends prompts, and any local PNG/JPEG/WebP image file named as an edit or reference input, to the image endpoint the user configures; no third-party network calls, no shell commands, no stored credentials, no telemetry.

mr-pewpyModelsv1.1.0Updated 1 day ago

About

Model Forge local image-generation backend — routes Hermes image_generate to any OpenAI-compatible images API (media gateway, vLLM-Omni, SGLang Diffusion, ComfyUI). Companion package for the Model Forge plugin. Disclosure — sends prompts, and any local PNG/JPEG/WebP image file named as an edit or reference input, to the image endpoint the user configures; no third-party network calls, no shell commands, no stored credentials, no telemetry.

This plugin has no README in the catalog yet.

More by mr-pewpy

5 plugins by mr-pewpy in the catalog.

mr-pewpymodel-forgeLocal models for every modality — chat, vision, OCR, image generation and editing, video, 3D meshes, music, SFX and voice. Model Forge finds your hardware (local probes; remote compute targets use the budgets you declare), shows which models fit, maps each model to the right inference engine (llama.cpp, Strata, Ollama, vLLM, ComfyUI, a user-configured OpenAI-compatible media gateway and more), pulls the model, and wires it into Hermes as a provider — so chat, image_generate, video_generate, vision and TTS run on hardware you control with no account and no API key. Six tools: model_forge_status (targets, VRAM/RAM, engines, loaded models), model_forge_browse (per-modality model catalog with hardware-fit verdicts), model_forge_target_add (register compute targets), model_forge_install (pull models through engine-native paths), model_forge_activate (wire a running chat model into Hermes), model_forge_generate_media (3D, music and SFX jobs through a user-configured OpenAI-compatible media gateway). Disclosure — model pulls download weights from huggingface.co or modelscope.cn on user request, and model_forge_browse queries the public huggingface.co model API for fresh listings unless fresh=false (optional HF_TOKEN / MODELSCOPE_API_TOKEN pull credentials live in plugin settings, are read only through Hermes' secret route and are sent only to their own host family); shell commands are read-only hardware probes (nvidia-smi, rocm-smi) plus vendor engine installers — model_forge_install action=engine_install runs a pinned vendor installer (a SHA-256-verified install script, or a pinned git commit checked out and re-verified before setup runs) in a background job only after Hermes approval; every child process (installers, engine servers, hardware probes) runs with a minimal environment (PATH, HOME, LANG, TMPDIR plus declared engine variables) that never carries credentials; hardware probes read local system state (GPU queries, sysfs, the Windows memory API) while remote compute targets use user-declared budgets; adding a compute target, model_forge_activate with apply=true (which rewrites model.default, model.provider and base_url in config.yaml), running a vendor installer, and starting an inference engine server process each ask for Hermes approval first; started engine servers (llama-server, vLLM) bind 127.0.0.1 only; media jobs and Maya pulls POST prompts and image bytes to user-configured endpoints; the bundled TTS helper POSTs text and base64 voice-clone reference audio to a user-configured /v1/audio/speech endpoint; background processes are the download daemon thread, approved installer jobs and approved engine server processes; no telemetry.Communitytoolsmr-pewpymodel-forge-desktopModel Forge cockpit pane for Hermes Desktop — compute targets, live engines, loaded models and hardware at a glance. Companion UI for the Model Forge plugin. Disclosure — the pane polls each compute target you register in Model Forge (plus http://127.0.0.1:8080 by default) with read-only GET requests to /v1/models, /api/tags and /system_stats to show which local engines are running, reads Model Forge's own targets file through a read-only shell command, and runs Model Forge's read-only hardware and memory probes when you open the resource view; no generation calls, no writes, no credentials, no background processes, no telemetry. Disclosure — model pulls download weights from huggingface.co or modelscope.cn on user request, and model_forge_browse queries the public huggingface.co model API for fresh listings unless fresh=false (optional HF_TOKEN / MODELSCOPE_API_TOKEN pull credentials live in plugin settings, are read only through Hermes' secret route and are sent only to their own host family); shell commands are read-only hardware probes (nvidia-smi, rocm-smi) plus vendor engine installers — model_forge_install action=engine_install runs a pinned vendor installer (a SHA-256-verified install script, or a pinned git commit checked out and re-verified before setup runs) in a background job only after Hermes approval; every child process (installers, engine servers, hardware probes) runs with a minimal environment (PATH, HOME, LANG, TMPDIR plus declared engine variables) that never carries credentials; hardware probes read local system state (GPU queries, sysfs, the Windows memory API) while remote compute targets use user-declared budgets; adding a compute target, deleting a model-storage entry, model_forge_activate with apply=true (which rewrites model.default, model.provider and base_url in config.yaml and adds a custom_providers entry), running a vendor installer, and any broker claim, swap, release or media job that would start an inference engine server process each ask for Hermes approval first; started engine servers (llama-server, vLLM) bind 127.0.0.1 only; media jobs and Maya pulls POST prompts and image bytes to user-configured endpoints; the bundled TTS helper POSTs text and base64 voice-clone reference audio to a user-configured /v1/audio/speech endpoint; background processes are the download daemon thread, approved installer jobs and approved engine server processes; no telemetry.CommunityDesktop appmr-pewpymodel-forge-providersLocal engine model-provider profiles for Model Forge (Strata, Project Maya) — run Hermes chat on local MoE engines and make their models selectable in /model. Companion package for the Model Forge plugin. Disclosure — sends prompts and requests to user-configured endpoints; no third-party network calls, no shell commands, no stored credentials, no telemetry.Communitymodelsmr-pewpymodel-forge-videoModel Forge local video-generation backend — routes Hermes video_generate to any OpenAI-compatible video job API (media gateway, engines with /v1/videos). Companion package for the Model Forge plugin. Disclosure — submits video jobs to user-configured endpoints; no third-party network calls, no shell commands, no stored credentials, no telemetry.Communitymodels

Developer Mode

Explore built-in Hermes toolsets and actions.

HermesAgent

Developer?

Create a Hermes extension for Desktop or agent workflows.

Developer Section