Item 3 / Meshly CLI

Your coding agent stays in the building.

You work with a coding agent in your terminal, on a model you run yourself. Code, prompts and answers all stay inside your network.

Meshly CLI is for teams whose code cannot be sent to a public AI service. It talks to your own model server and reports to your own Meshly Build instance, so its work is planned, reviewed and approved the same way as any other agent’s.

SYSTEM / AGENT OPERATIONS LAYER4 LAYERS

Four layers, from top to bottom. Agents and models is highlighted.

  • 4 · Agents and models (highlighted)Coding · Process · Local · CLIReplaceable workers, from any vendor.
  • 3 · Meshly BuildRules · Review · RecordNo agent approves its own work.
  • 2 · Meshly Data StackYour data · Meaning · PermissionAgents see only what they are allowed to.
  • 1 · Your infrastructureOn-prem · Private cloud · Airgapped
Meshly CLI is one of the agents, on a model you run.FIG. 1
Terminal agentLocal modelmacOS · Windows · Linux

How it works

The workflow stays the same, and only the model changes.

  1. 01 / Step

    It picks up a task.

    The agent takes its task from the Build board, just as Claude Code or Codex would.

  2. 02 / Step

    It works in a sandbox.

    It reads and changes code inside a sandbox, and it asks first before any action that needs permission.

  3. 03 / Step

    It thinks on your model.

    Every model call goes to a server you run, and nothing is sent to a public AI service.

  4. 04 / Step

    It hands the work in for review.

    The change returns to Build, where gates, audit and cost accounting work as usual.

Section A-A, network boundary

You can run it inside an airgap.

In an airgapped environment the whole set is local: Meshly CLI on the developer machine, your model server, and your own Meshly Build instance. Meshly Data Stack can run in the same network.

Specification

Here is what Meshly CLI is.

Type
Coding agent for the terminalIt combines an agent loop, a permission gate and a sandbox.
Model
Any OpenAI-compatible endpoint you runThat can be your own vLLM server or the Meshly AI appliance, for example.
Board
Connects to your own Meshly Build instanceIt uses the same MCP connection that Claude Code and Codex use.
Platforms
macOS, Windows and Linux
Availability
By requestWe deliver it through the Meshly Customer & Partner Portal.

General notes

A few things are worth knowing.

  1. Meshly AI, built per customer. The Meshly AI appliance serves local coder and reasoning models on your own hardware behind one OpenAI-compatible gateway, and Meshly CLI is the agent that talks to it. We build and configure it for each customer who needs it.

  2. Hardware matters. A local model is only as capable as the hardware it runs on. Reference hardware exists for the appliance, and other configurations need tuning, so ask us before you assume yours fits.

  3. Governance does not change. Build sees the same kind of agent either way, so approvals, the audit trail and cost accounting work exactly as they do with a hosted model.

One level deeper

You can check the boundary yourself.

The Trust Center covers deployment boundaries, security practices, and what is confirmed and what is still in progress.

Open the Trust Center

Questions: hello@meshly.fi