Run open-source AI across all your machines. No cloud service. Your data stays on your own hardware and network.
Most AI tools run on one machine at a time. vimin-core changes that.
Most homes and offices have more computing power sitting idle than they realise. vimin-core treats all of it as one: send a task to every machine at once, each one runs it locally with its own data, and results come back in seconds. You decide what gets sent back to the center and what stays on the device.
With vimin enterprise
In the enterprise version, agents talk to each other rather than just receiving tasks. One machine can break a problem into pieces and hand them off to others. You can route tasks to specific devices, set rules about what data moves where, and plug the whole thing into your existing tools and workflows.
Architecture
One machine routes the work. The rest run the models.
Run vimin-core start-center on any machine in your network. It routes tasks. No inference happens here.
Run vimin-core start-agent on each machine. Each one loads its own models and registers with the center. No shared storage needed.
POST to /api/broadcast. The center routes work to agents, collects results, and returns them. All on your network.
# 1. Install (pick your hardware) pip install "vimin-core[mlx] @ git+https://github.com/pberlizov/vimin-public.git" # Apple Silicon pip install "vimin-core[llamacpp] @ git+https://github.com/pberlizov/vimin-public.git" # Linux / Windows / CUDA / CPU # 2. Start the center node vimin-core start-center # localhost only (single machine) vimin-core start-center --host 0.0.0.0 # accept agents from other machines # 3. Connect agents (run on each inference machine) vimin-core start-agent # same machine as center vimin-core start-agent --center http://<center-ip>:8080 # remote machine # 4. Broadcast a task vimin-core broadcast "Summarize Q3 results." --mode return
Enterprise
Nodes that talk to each other and run multi-step jobs without a human in the loop.
Nodes can break a task into pieces and pass them to other nodes on the network. Each step runs locally. OpenClaw nodes join the same fleet alongside standard agents.
Chain inference tasks across nodes so the output of one model becomes the input to the next. Define multi-step pipelines with conditional branching and per-step hardware routing.
Assign department-level administrative nodes that coordinate their own cluster of inference machines. Any graph topology works.
Choose exactly what flows back to the center node. Sensitive inference results can stay on the edge device entirely.
Connect vimin to your existing stack. First-party integrations include LiveKit for real-time audio and video inference pipelines. Build your own with the plugin API.
Run fine-tuned or proprietary models alongside standard registry models. Hardware backends not covered by vimin-core can be added with dedicated integration work.
Direct access to the engineering team. Response time SLAs with guaranteed escalation for production incidents.
Open-source vs. Enterprise
Start free and self-hosted. Upgrade when you need more.
Book a 15-minute call to talk through your deployment.