Private
AI
Mesh.
Federated RAG across machines you own. Local LLMs via Ollama. Tailscale mesh networking. Nothing reaches the cloud.
private by design
cloud API latency
to add a node
The Platform
Three layers.
One mesh.
Every component is self-hostable, composable, and talks only over your private Tailscale network.
Orchestrator
Central brainPreprocesses queries with llama3.2:3b, fans out to all Aligner nodes, reranks merged chunks, synthesizes final responses via llama3.1:8b streaming. JWT auth + RBAC.
Aligner
Edge RAG nodeIndexes your local files, serves security-filtered vector retrieval. Security levels enforced at source — the orchestrator never sees restricted chunks.
Client layer
Web UI + Tauri appBrowser WebSocket chat and a Tauri desktop dashboard. Connects to your orchestrator — no intermediary, no third-party relay.
How it works
Three steps to
your private mesh.
Install & pull
Run the orchestrator on any machine. Pull your Ollama models. Zero cloud configuration required.
Connect via Tailscale
Add nodes to the mesh with a single auth key. Encrypted WireGuard tunnels, no port forwarding.
Query your swarm
Send a prompt. It fans out to every node, reranks, synthesizes — streamed back in real time.
Capabilities
Everything
in one mesh.
Federated RAG
Fan-out to every Aligner node simultaneously. Merged, reranked results — zero data leaves your machines.
Security tiers
Per-file, per-line access control enforced at the edge. The orchestrator never receives restricted data.
Local inference
llama3.1:8b on your hardware via Ollama. Zero API calls, zero data leaks, zero latency tax.
Tailscale mesh
Every node auto-connected over zero-trust WireGuard. New machines join with one command.
Streaming chat
Token streaming over WebSocket with interactive node-selection canvas. Pick exactly which nodes respond.
Voice input
Hold-to-record. Whisper transcription runs server-side — query your mesh hands-free.
Live demo
A real query.
Your machines.
Watch the orchestrator fan out to three nodes, rerank the results, and stream a synthesized answer — all on local hardware.
Download
Get the desktop app.
One installer sets up the Swarm app and the local Aligner component. GPU acceleration is auto-detected — no flags needed.
$ curl -fsSL https://raw.githubusercontent.com/SwayamDani/swarm-releases/main/install.sh | bashx86_64 · beta · see the docs for install flags
Private by design
No cloud API ever called
Cloud overhead
Everything runs on your hardware
Starter tier
Up to 15 on Growth plan
To join the mesh
Via Tailscale auth key
Pricing