Meshly AI / On-prem appliance / Built per customer

Your own AI runs on your own hardware.

Meshly AI is a compact on-prem appliance that runs local models in three modes: reasoning, coder and artist. Your data, prompts and models never leave the box.

The artist mode covers image and video generation, and nothing calls a public AI service. You can reach the box in two ways: through Meshly CLI in the terminal, or through the Ilves chat app on desktop and mobile.

Meshly AI is built and configured for each customer who needs it, and it is in use where it fits. It is not a public download.

On-premAirgap-capableBuilt per customer

Specification

Here is what the box is.

On-prem
It runs entirely on your hardware.
Airgap-capable
It makes no calls to any public AI service.
Unified memory
Unified memory lets large models run on a single box.
FP4 inference
It uses modern low-precision acceleration.
OpenAI-compatible
One local API serves every tool.
Model-neutral
You can bring your own open or local models.
Versions
Meshly AI appliance 0.4.1, Meshly CLI v0.8.1, Ilves chat v0.4.2We measured these on 2026-11-18, and the appliance on 2026-08-20. In use, and far newer than Meshly Build.
Availability
Built per customer

Three modes

It reasons, it codes, and its artist makes moving pictures too.

One box gives you three modes.

You can switch between a careful reasoning model, a coder that runs the work, and an artist that generates both stills and video. The box runs one heavy model at a time and re-routes on demand, so the whole appliance points at the job in front of it.

ModeWhat it does
CoderIt is a coding brain that runs the work. It writes, edits and runs your code, and it drives Meshly Build and Meshly Data Stack for you, even fully airgapped.
ReasoningIt does heavy, careful thinking for triage, analysis and knowledge work, where depth matters more than speed.
ArtistIt generates images and video on the box, from prompt to image, prompt to video and image to video. No frame goes to the cloud.

Your own AI, your own stack

In coder mode, it runs Meshly Build and Meshly Data Stack.

The coder brain does more than write code, because it can operate Meshly Build and Meshly Data Stack for you, even fully airgapped. It moves tasks, queries data and runs the agent loop. Your own AI drives your own stack, and every call stays inside the perimeter.

  1. Operates Meshly Build. It plans, delegates, moves tasks and pushes results. In an airgap, it works against your local Meshly Build instance.

  2. Queries Meshly Data Stack. It queries the data through tools on the box and makes no external calls.

  3. An MCP server in its own right. Any MCP client can reach the box's image and video tools with nothing to install.

  4. Works fully airgapped. External connections on the box are off by default.

Two interfaces

You can reach it through Meshly CLI or the Ilves chat app.

You can talk to the box in whichever way suits the work, and the same appliance sits underneath both. Ilves is the name of the AI that the appliance hosts and that Meshly CLI drives.

Terminal
Meshly CLIThis is the agent in your terminal, for coding and for operating your stack.
Desktop and mobile
The Ilves chat appIt brings together Chat, Image Studio and Video Studio. It runs on macOS, Windows and Linux on the desktop, and on Android and iOS on mobile.
Models
Model-neutralYou can bring your own open or local models, and the box has no per-seat or per-token metering.

Meshly CLI

This agent CLI works fully offline.

Meshly CLI is the terminal interface to the box. You can point it at your appliance, or at any OpenAI-compatible endpoint your team runs, such as your own vLLM server. Engineers keep the habits of a modern coding agent, and the model stays in-house.

Claude Code and Codex call a cloud model through a public API. Meshly CLI calls Meshly AI on your own premises, with the same workflow and the same tools.

Same workflow, same tools

Read, Edit, Write, Glob and Grep tools
Bash execution with permission modes
MCP servers for extension
Plan mode and approval gates
Task, agent and subagent patterns
Session persistence and audit logging
  1. Same tool surface as Claude Code. Meshly CLI offers the same tool surface as Claude Code: Read, Edit, Write, Bash, Glob and Grep, plan mode and MCP.

  2. Workflow-compatible. Meshly CLI matches the workflow of these tools and does not use their protocol, so no public AI API is in the picture.

  3. Standalone or connected. You can use it standalone, or connect it to Meshly Build to dispatch tasks and stream results back. In an airgap, it connects to your local Meshly Build instance.

Airgapped by design

The box keeps every connection inside your network.

Inference stays on the machine. There is no telemetry, no auto-update ping and no anonymous usage statistics. On a shipped appliance, external connections are off by default, and you opt in per box with explicit allow-lists. You can run it on a closed network.

  1. Inference stays on the box. The box has no cloud fallback, ever.

  2. No telemetry or phone-home. The only traffic is the traffic you configure.

  3. Your git server can be on the box too. When it is, the code has nowhere it needs to go.

  4. Airgap-verified. Models are pre-cached, and the box runs with the network unplugged.

Who needs this

It is built for teams that keep their data in-house.

Meshly AI brings the model to the data, and the whole loop runs on a box you own.

Cloud AI works well until your compliance officer reads the data processing addendum.

SectorConstraint
Defence and governmentThese are classified networks, where code and prompts cannot leave the perimeter.
Finance and bankingData sovereignty applies, outbound network policies are strict, and the work is audit-bound.
HealthcarePHI must not reach cloud AI, and environments are HIPAA-bound.
Regulated R&DThe work involves pre-release IP, patent-pending code and legally sensitive material.

Availability

Built for each customer who needs it.

Meshly AI is not a public download. Tell us about your environment: what hardware you have, which models you run and which restrictions you operate under. We will work out whether it fits and get you a box with the right configuration.

See also

One level deeper

You can run the whole set in one airgap.

Meshly Build, Meshly CLI, Meshly Data Stack and the Meshly AI appliance can all run on your own hardware, and the Sovereign page states the maturity gap plainly.

Open Sovereign