Distributed inference / control plane
One API.
Many machines.
Dendrite gives agents and applications one clean OpenAI-compatible surface for inference routed across Linux, CUDA, Windows, WSL2, CPU, and GPU providers.
Network topology
One control plane · distributed execution
Start here / provider utility
Download once. Double-click. See what this machine can do.
The Dendrite Provider Forge opens a private dashboard on your own machine. It discovers hardware, recommends compatible GPUs, installs a verified runtime, downloads approved weights, and guides enrollment without making you chase dependencies.
guided and checksum-verified
Provider onboarding
One app takes the machine from scan to serving
Double-clicking the download opens an actionable dashboard on 127.0.0.1. The cloud page never receives local setup credentials.
The released app needs no Node, npm, Python, or source checkout.
Choose the platform above. Checksums and a machine-readable release manifest are published beside every binary.
The app scans CPU, RAM, GPUs, VRAM, PCIe, ISA, disk, network, and existing engines, then opens Provider Forge in your browser on a random 127.0.0.1 port.
Double-click dendrite-windows-x64.exeThe dashboard recommends a compatible serving GPU set, installs the pinned llama.cpp runtime, and downloads only models that fit. A Quadro can stay installed while an RTX card serves.
Paste the operator invite into the local dashboard. Enrollment returns a provider-scoped connection credential directly to the app and keeps it in memory. The app starts the engine and shows connection status.
Network pulse
Capacity matching
Consumers state what they need; the coordinator selects an eligible provider automatically. Provider inventory is optional operator diagnostics.
Operator console
Send a completion
A lightweight client for testing the same API agents use.