Agents Node

AI that runs on your machines.

The SweetHive Agents Node is a desktop app you install on your own hardware. It runs your context agents with local models, plus downloadable render workers for heavy AI & GPU jobs. Nothing sensitive leaves your machine unless you publish it. This is what lets SweetReach turn a brief into finished marketing videos and attributed leads.

The idea

Your hardware, inside SweetHive's boundaries.

Context agents, local models

Connect an agent to a context; it answers from what that context's groups can see, using a local model (Ollama) or Claude: on your node, not in a third-party cloud.

Downloadable workers

Add capabilities on demand: install a HeyGen worker (cloud avatars, no GPU) or a local GPU worker straight from the app. The node picks up jobs and runs them with the worker you chose.

Private by construction

Agents can be pinned to your on-prem node, with scope enforced down to the hardware. GPU rendering runs locally; only the finished asset is uploaded to the cloud.

Use case · SweetReach

From a one-line brief to attributed leads.

SweetReach is SweetHive's internal marketing engine. The heavy lifting (writing the script with the context's agent, rendering the video on a GPU or via HeyGen) happens on your Agents Node. The cloud only orchestrates and stores. Here is the whole loop:

Brief one line: what to say, tone Script the context AGENT writes it, on a node, grounded in the context Render HeyGen or GPU worker ON YOUR NODE → local file → cloud upload Publish YouTube · LinkedIn Meta · TikTok · X with tracked links Leads captured & managed in SweetCRM attribution: which video/campaign produced each lead → informs the next brief

The two AI steps are deliberately split. Script is an AI task, so it runs on the agent connected to the campaign's context, which can live on a different node/PC, with its own model. Render is a hardware task, so it runs on the node that has the GPU worker (or the HeyGen key). One workflow, the right machine for each job.

Workers

Download a worker, get a capability.

Rendering is pluggable. Open the Node app, pick a worker, install it, and the node starts claiming video jobs and rendering them your way.

Cloud

HeyGen worker

Photoreal AI avatars via the HeyGen API: no GPU required, just a key. The node submits the script, downloads the finished MP4 locally, then uploads it to SweetHive storage.

GPU

Local avatar worker

A talking-head avatar rendered entirely on your GPU, offline, with a local voice. Best on a workstation or server; nothing but the final file ever leaves the machine.

More

Bring your own

The worker interface is small: take a job, return a video file. New workers publish to the catalog and appear in the app, no app update needed.

The workflow is in the cloud. The heavy AI is on your metal.

Run SweetReach, and any agent, on hardware you control, inside the same context boundaries that make SweetHive safe.

Download the Agents Node Get a demo