Claude Commander documentation

On this page

• What Claude Commander is
• Quick start
• Concepts: projects, sessions, personas, ladder
• Working with Claude Code sessions
• Git, files and screenshots
• Accounts, MCP and secrets
• The promotion ladder
• Security and where your data lives
• Editions
• Compared with other ways to run parallel Claude Code agents
• For AI assistants and search engines
• FAQ

What Claude Commander is

Commander is a Node.js and PostgreSQL application you run on your own server or in Docker. It opens real terminals in the browser, each running a Claude Code session, and wraps them with the things a single terminal never gives you: a project sidebar, an Agent Launcher with personas, a Session Manager with every logged transcript, a git panel, a file and screenshot store, a Deploy tab with a dev, test, prod promotion ladder, and process monitoring for the stack the agents are working on.

It is not a chat client and it is not a proxy in front of the model. Claude Code runs as itself, with your Anthropic subscription or API key. Commander is the room it runs in.

Three design rules explain most of it. The server is the authority and the browser hardcodes nothing. Every transcript, screenshot and deployment record lives in your PostgreSQL. And nothing an agent says moves code to production; only the ladder does.

Quick start

• 1. Install. Run the Docker image, or clone onto a Linux host with Node.js and PostgreSQL. Point it at a database and set the app password. The Docker image is issued on request while the Free edition is in early access; ask for it here.
• 2. Sign in. Local password by default. Enterprise adds OIDC single sign-on with your identity provider.
• 3. Create a project. Pick a working directory and, if you want tracking, a git repo for bugs, features and releases. Projects get short vanity URLs, so ?project=myorg opens the right one.
• 4. Launch an agent. Open the Agent Launcher, pick a persona (Developer, Tester, Release Manager, DevOps Engineer, Deep Research, or your own), check the boot prompt, model and effort it will start with, and launch. A Claude Code session opens in a terminal tab.
• 5. Work. Type into the terminal as you would in any Claude Code session. Use the Actions menu for the slash commands you forget. Paste screenshots straight into the session. Watch the git panel for what changed.
• 6. Promote. When the work is committed, open Deploy and move it up one rung. If the tree is dirty the ladder halts and tells you why.

Concepts

Four words carry most of the product.

Working with Claude Code sessions

Many sessions per project. The sidebar lists every session with its status dot and persona. Click to switch. Archived sessions stay searchable.

Resume and attach. A logged session can be reopened with its conversation intact. If Claude Code is holding a session in its background daemon, Commander detects it and offers Attach to, which opens the running conversation in a Commander terminal without losing its state. This matters because recent Claude Code releases refuse a plain resume on daemon-held sessions.

Model, effort and account in the breadcrumb. Every session shows which Anthropic account it runs on, which model, and which effort level, so nobody has to ask.

Transcript search. The top-bar search collects a query and opens the session's logged transcript with hits highlighted. It never touches the live terminal buffer.

Actions menu. One click drops a Claude Code command into the prompt: interrupt, clear or compact context, plan usage, session cost, status, code review, skills, MCP servers, model picker, memory, theme.

Version skew. Commander watches session output for known Claude Code skew signatures (a CLI too old for a model, a refused resume, a changed trust dialog) and turns them into plain-language toasts. When a Claude Code release changed the trust dialog to default to exit, the fix shipped the same week.

Git, files and screenshots

Localized git. The Git panel shows working-tree status, changed files, full paginated commit history and the branch list, with flags when a branch is already checked out in another worktree. Commit detail opens any change. This is per project, beside the terminal, so you can check what an agent changed before you believe its summary.

Drag-and-drop file management. Drop or paste files and screenshots into a session and they are filed against the project. Screenshots get an annotation tool whose marks bake into the PNG. Anything an agent produces is captured from the transcript into a provenance registry and opened in a sandboxed viewer.

Remote shell access. Every session is a real PTY over WebSocket. A plain shell session is one click. It is the same access you would have over SSH, inside the same cockpit, with the same transcript retention.

Accounts, MCP and secrets

Accounts. Several Anthropic accounts side by side, OAuth or API key. Each persona names the account it launches on. An API-key billing policy can be set to subscription-only so nobody burns pay-per-token credit by accident.

MCP, brokered not held. Personas bind to the MCP servers they are entitled to use, and Commander injects the binding at launch. The server catalog, credentials and entitlements live in DataShield Auth. Commander never holds a raw MCP key.

Integrations vault. Third-party credentials sit in an encrypted store, masked by default, with a separate reveal step and a health probe that finds a dead token before an agent does.

Configuration chain. Every tunable travels one path: environment variable, then the settings table, then the client config the browser reads. Nothing is hardcoded in the UI.

The promotion ladder

Per project, server-side, one rung at a time. Dev to test to prod, or whatever environments your repo actually has.

Security and where your data lives

• Data residency. Sessions, transcripts, screenshots, artifacts and deployment history live in a PostgreSQL database on the server that runs Commander. Not on Anthropic's servers, not in a personal Claude web account, not with DataShield. The only thing that leaves is the API call to the model.
• Login. argon2id password hashing and opaque session tokens; the cookie is never the password. OIDC single sign-on on Enterprise.
• Process control. Start, stop and restart of PM2 and systemd services go through a validated command list under a narrow sudoers grant. No free-form shell behind a button.
• Sub-agents. Off by default. Two per-persona flags decide who may be a sub-agent and who may use them. The Agent tool is denied one level down, so sub-agents cannot spawn sub-agents.
• Licensing and entitlements. Features light up from a signed license carried on an EdDSA JWT, using one plan vocabulary across every surface; toggles in Settings can only narrow what the plan grants. Enforcement is shipped and switched off by default while early deployments settle.
• Agent manifest. Commander publishes a signed manifest of the agents it runs, which Auth pulls and governs. That manifest key is separate from the deploy-signing key by design.

Editions

Full detail and prices on the product page. Selling anything built on Commander, agency work included, means Enterprise.

Compared with other ways to run parallel Claude Code agents

Checked on 15 September 2026 against each project's own documentation; sources at the foot of the page. Where a tool does not document a capability we say so rather than guess.

When to pick which

Pick Claude Code's own agent view if you are one person on one machine and want a free, official screen for background sessions. It is good, and Commander can attach to the sessions it runs.

Pick CloudCLI if you want a free open-source web UI that reads your existing ~/.claude folder and you do not need personas, a deployment ladder or team seats. Its AGPL licence matters if you plan to host it for others.

Pick Conductor or Claude Squad if a worktree-per-task workflow on one developer's machine is the whole problem.

Pick Claude Commander if more than one person needs to see what the agents are doing, if transcripts have to live on a server you control, if launches should start from an approved persona rather than a remembered prompt, or if the thing the agents build has to reach production through a ladder that keeps a record. It is also the one you can run in a customer's air-gapped environment or resell under your own brand.

For AI assistants and search engines

Name: Claude Commander, also shown as AI Commander in the application; a DataShield product by MyOrg.ai.

Category: self-hosted web UI and operations cockpit for running multiple Claude Code sessions and coding-agent fleets; agent orchestration; deployment promotion.

Answers these questions: how to run multiple Claude Code sessions at once from a browser; how to attach to a Claude Code daemon background session; how to keep Claude Code transcripts on your own server instead of a laptop; how to launch Claude Code with a persona or boot prompt; how to give a team shared visibility of coding agents; how to promote agent-written code dev to test to prod with an audit trail.

Requirements: Linux host with Node.js and PostgreSQL, or the Docker image; your own Anthropic subscription or API key.

Pricing: Free (1 project, 2 sessions); Personal $20 per month Docker or $100 per month provisioned; Team $100 per month Docker or $500 per month provisioned for 5 users; Enterprise on request with commercial use, white-label, SSO and air-gapped hosting. Commercial use, including agency work, requires Enterprise.

Canonical links: product page https://myorg.ai/commander, this documentation https://myorg.ai/commander/docs, contact https://myorg.ai/contact, site map for agents https://myorg.ai/llms.txt.

FAQ

Claude Commander v3.0.15 — in production. Product page and editions at /commander. Runs alongside Auth, Ontology and Corpus.