Meshly Build / Station

Station is the desktop runner, native to your machine.

Station runs Claude Code and Codex agents on your workstation, and every task gets its own isolated workspace.

Station lands approved work and streams agent output back to the dashboard. Architecture scanning runs locally, so the scan never sends your source anywhere. Station is a native app for macOS, Windows and Linux.

Desktop runnerVersion 1.7.0Claude Code · CodexmacOS · Windows · Linux

Claude Code and Codex

One runner handles both agents.

Station runs both Claude Code and Codex on macOS, Linux and Windows, so every engineer works in the agent they are fastest in, for writing code and for reviewing it.

Station launches each agent with the model, effort and speed you chose for it in Build, and every running session shows which runtime it is on.

Claude Code
Station runs Claude Code for coding and review sessions.Each session uses the model, effort and speed set for its agent
Codex
Codex works on the same board and follows the same workflow.It uses the same per-agent model settings
Your plugins
Plugins are chosen per project.Each project chooses whether sessions load the plugins from your own installation
Platforms
Both runtimes run on macOS, Linux and Windows.

Isolated workspaces

Every task runs in its own workspace.

Each task gets its own isolated copy of the repository, so parallel work never collides, a review always reads the exact commit it judges, and your own checkout stays yours.

A session that writes code is sandboxed. On macOS and Linux it cannot read your credential stores, Station’s own data or the checkouts of unrelated projects, and on Windows it cannot reach Station’s data. Agents can use podman and docker without your registry credentials.

Workspace
Each task has its own copy of the repository.Parallel tasks never share a checkout
Review
A review runs at the exact commit.Checks and the security scan run against the work under review
Landing
Station lands work on the task’s target branch, as a merge or as a pull request.The work lands exactly as it was approved, once it is approved and Build accepts the landing
Merge tracking
Station recognises merges wherever they happen.This way the board knows what shipped

Per-project participation

Choose how each Station takes part.

One setting per project, per Station, decides how this machine takes part, so every engineer’s machine contributes exactly as much as it should.

Cursor and GitHub Copilot get configured against the same checkout.

Automatic
Build starts sessions on this Station.
Manual only
A person drives the work.The work still flows through the board, the review and the record
Off
This machine sits the project out.

Native desktop app

Station runs as a desktop app on every platform.

Station is a small download with low overhead, and it gives you a tray icon and menu controls on every platform.

PlatformBuildInstallerIncludes
macOS13+Universal binary, Intel and Apple Silicon.dmg installerTray menu, native app
Windows10 / 11Native app.msi and .exe installersTray app, headless mode
Linuxx86_64Native buildNative Linux buildTray app, headless mode

The tray menu offers Show / Hide, Pause All, Resume All, Refresh MCP, Settings and Quit.

Architecture scanner

Station scans locally with no extra setup.

The architecture scanner runs in-process inside Station, so you install no separate tool and no extra toolchain, and no external service takes part.

The scan walks your repository, detects surfaces and builds an envelope with a fixed schema. Only the envelope uploads to Server for the team, and the scan sends none of your source.

The envelope carries repo_sha, scan_id and timestamps, which keeps the bookkeeping reproducible. You can regenerate it on demand, and Server stores the result so the whole team sees the same thing.

Scan stepHow
Walk the source treeIn-process
Detect surfacesNative pipeline
Extract dependenciesNative pipeline
Build the envelopeFixed schema
Upload to ServerAuthenticated

Live streaming

Watch the agent work, from any browser.

Agent output streams from the workstation to the dashboard in real time. The full Claude Code or Codex session, with its edits, test runs and decisions, is visible to anyone with access to the project.

What streams
Station streams stdout, file edits and tool calls.The stream is live, so the dashboard never has to poll for updates
Resilience
The stream survives laptop sleep, network blips and server restarts.Streams recover across reconnects, sleep / wake and deploy restarts
Who can watch
Anyone with access to the project can watch.Product and engineering managers can watch without installing anything as an operator

Command palette

Press ⌘K to reach everything.

The command palette opens on ⌘K and lets you run tasks, switch projects and move between views. Search is fuzzy and keyboard-first, and it covers tasks, projects and recent actions.

Palette
The palette runs tasks, switches projects and opens views.It opens on ⌘K
Settings
Settings sit in three tabs: General, Projects and About.Nothing is buried two levels deep
Sidebar
The sidebar holds Home, Agents, Projects and Logs.

Headless reliability

Station runs unattended and recovers on its own.

Headless agents run to completion unattended and recover cleanly from sleep / wake, so long-running CI usage works as well as interactive operator usage.

Unattended sessions
Sessions run to completion.No confirmation prompt waits on an operator
Automatic reconnects
The live feed picks up again by itself.This happens after sleep / wake or a network blip
Session log retention
You set the retention period in days.You can keep session logs for as many days as you choose, or keep them all
Live log-level switching
Station writes structured logs.You can change the log level from the UI without restarting Station

Runners

Station is the GUI runner, one of two.

Server is the control plane, and runners connect to it.

PartWhat it is
ServerControl planeServer is the shared workspace for boards, agents, review, audit and knowledge. Meshly hosts it, or you host it on your own infrastructure.
StationGUI runner, desktopStation is a native desktop app on each operator’s workstation. It runs Claude Code and Codex agents, scans architecture and streams output, natively on macOS, Windows and Linux.
Meshly CLILocal-AI runner, airgappedMeshly CLI is a standalone coding agent for your own local model, and it connects to Build like any MCP-aware client.

General notes

Good to know.

  1. Getting Station. Tell us about your team, and we will walk through the Station rollout with your real environment, including how it fits your managed-software setup.

  2. Station and Server. Station is the agent runtime, and Server is the control plane it reports to, which Meshly or you can host.

  3. Governance. How agents are governed, from guardrails to approval, is covered on Control.

One level deeper

See the control plane.

Server holds the boards, agents, review, audit and knowledge, and it runs the MCP server that Station reports to.

Open Server