01
Messaging gateway
Receive and reply through Telegram, Discord, or a generic webhook from one process.
Explore channelsSelf-hosted AI control plane
One binary.Your models.Your rules.
Connect real messaging channels to local and cloud models, then decide how every conversation is routed. Korvun keeps the gateway, policies, and orchestration under your control.
Current releasev0.14.1 — BetaApache-2.0
The name
Kernel for Orchestrated Routing — Versatile Unified Nodes
Kernel
One binary at the center. The same kernel from a Raspberry Pi to the cloud.
Orchestrated Routing
Policies decide which model answers — where it runs, and how far your data travels.
Versatile Unified Nodes
Channels, brains, models and agents living under one roof, as one mesh.
The name is the logo. Signals enter, the kernel decides, one governed reply exits — the K has been drawing the acronym all along.
Verified delivery
Install with proof, not promises.
v0.14.1 — Beta publishes six headless OS/architecture archives. The release also includes a cosign-signed checksum manifest and an SBOM for every archive.
016 headless archives
02cosign-signed checksums
03SBOM per archive
Read the complete install and verification guide korvun · install
VERSION=0.14.1
curl -LO https://github.com/Sebastian197/korvun/releases/download/v${VERSION}/korvun_${VERSION}_linux_amd64.tar.gz
curl -LO https://github.com/Sebastian197/korvun/releases/download/v${VERSION}/checksums.txt
sha256sum --check checksums.txt --ignore-missing
tar -xzf korvun_${VERSION}_linux_amd64.tar.gz
./korvun version✓ SHA-256✓ cosign✓ SBOM
Product map
Eight pillars. One binary.
Each part solves a concrete step between an incoming message and a governed model response.
02
Multi-model routing
Place local Ollama and cloud Groq models behind the same routing layer.
Understand the model03
Multi-brain orchestration
Give each brain its own models, dispatch policy, and persona.
See the architecture04
Dispatch policies
Configure how eligible models run and how Korvun chooses the response.
Open configuration05
Visual builder
Compose channels, brains, models, and validated connections on a canvas.
Meet the builder06
Korvun Desktop
Run the gateway and builder from the packaged desktop application.
Install Korvun07
Governed memory
Brains store bounded notes through a governed tool and recall a session deliberately — nothing is remembered by accident.
Meet governed memory08
Signed releases
Verify the checksum manifest and inspect the supplied SBOM before running a binary.
Browse releasesRouting you can inspect
Sensitive data never leaves the machine.
When a brain is marked private, cloud models are excluded from its eligible targets. The visual builder shows that exclusion as part of the graph.
Inspect the configuration referenceprivate brainLOCALlocal model
private brainCLOUD · excludedcloud model
privacy: local_only
Product proof
A cloud model, governed tools, one config.
Sixty seconds inside the Korvun desktop app: gpt-4o-mini — reached through the v0.9.0 universal gateway — is asked to remember something, the governance gate authorizes the memory tool, /notes shows the stored note, and the model answers from memory.
Build your first route
Run it today.
Start with one channel, one brain, and one model. The quickstart keeps the first configuration small and explicit.