Private mesh, zero config
Every machine gets a stable private identity and address. Peer-to-peer when possible, relayed when not — you never think about it.
Private AI infrastructure
Your AI.
Your machines.
Usable by your team.
Agent-Fabric turns the models, tools and agents running on your own hardware into private services your team — and your customers — can actually call. Encrypted mesh, scoped access, one command. No public ports. Ever.
demo-laptop — fabric
How it works
Three commands from laptop to team-ready.
-
01
Connect
Join this machine to your private network. WireGuard-grade encryption, NAT-punching, relay fallback — handled.
$ fabric up -
02
Publish
Turn a local model, MCP tool or internal API into a named private service. It never opens a public port.
$ fabric serve http://127.0.0.1:11434 --name my-model -
03
Share
Grant scoped, expiring access to teammates — or hand a customer a time-boxed demo you can revoke with one click.
$ fabric grant llm://my-model
See it run
The CLI, exactly as it behaves.
Real recordings of fabric — not mockups.
Two consoles, one system
Cloud manages trust. Local does the AI work.
The cloud console sees who is connected and what is shared — never your prompts. The local workbench detects hardware, publishes runtimes and runs agents on your machine.
Works with your stack
Plug in the tools you already run.
Agent-Fabric doesn't replace your runtimes, models or cloud — it makes them reachable across your private mesh. If it speaks a standard API, it publishes in one command.
Publish any runtime
Ollama
vLLM
llama.cpp
LM Studio
Any model & framework
Hugging Face
PyTorch
Llama- GGUF
Speak the standards
OpenAI API
MCP- A2A
WireGuard
On your hardware & cloud
NVIDIA
Apple
Linux
Docker
Python
AWS
Open CLI — install in one line, publish in the next.
Why teams pick Agent-Fabric
Built for AI that must stay yours.
No public ports
Services are reachable only inside your network. There is nothing to scan, nothing to firewall, nothing to leak.
Scoped capabilities
Access is a signed, expiring token for one action on one service — not a network hole. Revoke any of it instantly.
Time-boxed customer demos
Give a prospect real access to your private model for 7 days — monitored, revocable, and gone when the clock runs out.
Local-first AI workbench
Detect GPUs and runtimes, publish local models, and run agent loops on your own hardware from one console.
Signed & observable
Signed releases, signed network maps, full audit trail, and a metering pipeline you can bill on.
The data boundary
Your prompts never leave your machines.
One clear line splits the system. Above it, the cloud coordinates trust. Below it, your AI runs — and your prompts stay.
We couldn't read your prompts if we wanted to.
Make your private AI usable today.
$ curl -fsSL https://docs.falconoon.com/install.sh | sh