◧Five views
Focus one session, full size · Grid up to four live terminals · Board a tile per session · Review the diffs · History every conversation.
Open source · Bring your own keys · Local-first · MIT
Your agent lab.
Claude Code, Codex and your shells, side by side: isolated, observed, in your control. Launch an agent already set up for its project, watch every one of them, and give the risky work a branch, a sandbox or a machine of its own.
Every agent. Every environment. One control room.
What is in the lab
From the moment you launch an agent to the cloud machine it borrows. Every picture below is the real app, captured on a private virtual display with a fictional workspace.
01 · Session manager
Claude Code, Codex or a plain shell, in a real terminal that keeps running in tmux when you close the window. What Agent Manager adds is everything set up before the first prompt.
~/.ssh.Repository
am/<name>: ten agents share a repository without stepping on each other.In its own git branchPlanned Launch presets: keep a set of these choices under a name and start from it in one click. Today a group already carries its default permission level.
02 · Interface, views, themes
Not columns of text: pages drawn like instruments, a terminal of our own (Ghostty's core, drawn by Godot), five ways to look at your sessions and seven themes shared with Second Brain. Try them.
Focus one session, full size · Grid up to four live terminals · Board a tile per session · Review the diffs · History every conversation.
FocusTerminalNeonEveningForestMorningPaper
The same as Second Brain's. A click changes the whole window, the terminal's colours with it.
Ctrl+Shift+K the palette: every action and session, by name. With Ctrl+Shift: F G B H R the views, A approves the first agent waiting. Change any of them.
03 · The control room
Claude Code and Codex report what they do through their own hooks; Agent Manager turns it into lights you read from across the room. The ones that need you come first.
04 · Continuity
Every session keeps every conversation it ever had. Agents live in a private tmux server, so the app can close and the work goes on.


05 · Better together
Agents forget everything when their window closes. Agent Manager connects each one to Second Brain GD, a separate app of the GD projects, through MCP: the agent reads the project's memory and proposes what it learned, and the Brain's archivist files it, under your rules and with your OK. The next agent knows, and so do you.
Runs the agents. Gives each one its own Brain key at launch, says how to document the work, and keeps each session's summary.
Keeps the project: vision, decisions, the owner's words, where things stand. Its archivist files every proposal with your rules, and you see who proposed what.
git log.
06 · Built-in MCP
Every Claude Code or Codex session is offered Agent Manager's own MCP server, nothing to install. A lead agent looks at any session and starts helpers of its own: they show up in your list, « ↳ Lead », in real terminals you can take over at any time.
Reads · your sessions
Writes · its own children
The helpers a lead starts appear under it, « ↳ Lead », each in its own terminal: watch them work, take one over, or stop it. Nothing happens off screen.
◐ Connect other tools
The same server for any MCP client outside the app, with the command to paste: Claude Code, Codex or another tool can list, read and start sessions. It can never type into one, nor approve.
Built on it · Orchestrator beta
A Claude Code session with an orchestrator's instructions and a few more tools
(list_repos, ask_owner, finish_session), for all your connected
repositories. It reads your Second Brain, proposes a plan and waits for your OK, starts each task in its own
worktree with the tool and model your rules choose, and hands you every result to review. It never pushes
nor merges: you do.



07 · Repositories and keys
Connect a repository on GitHub or Azure DevOps: Agent Manager makes an SSH key for it, checks
it, and gives it to that repository's sessions only. No ~/.ssh to edit, no key shared between
projects.
GIT_SSH_COMMAND, with its own known_hosts: nothing written to your SSH config.
How to read a line
08 · Your machine
The ground button's page reads this computer and WSL: memory, processors, the true room on the disk, and the memory each session holds, from the last hour to the last seven days. Hover a bar, and it tells you who weighs what.
The button in the top bar is alive, and its tooltip says why. On Linux it reads « Computer ».
09 · Cloud
Give a risky task its own Fly.io machine, with your own Fly token. Pick a size and see Fly.io's published price before you deploy; then watch its health and bring its ports home through one SSH connection. When a service there is ready, bind one of its ports to the machine's public URL; unbind it in one click.
the sandbox's host key is checked · nothing is made public · a paused sandbox is never started
Filmed with stand-ins for Fly.io and SSH: nothing was deployed, nothing was billed. The prices are Fly.io's published ones, as the app shows them. The public URL was tried once on Fly.io itself, on a throw-away machine.
10 · Voice and mobile
Hold F8 and speak: your words appear in a see-through box, phrase by phrase, and stay there when you release the key. Read, correct, add; Enter sends all of it, and nothing before.
Into a box you send with Enter. With your OpenAI key, about $0.003 a minute, shown.
« approve Flaky tests », « who is waiting? », « grid view ». Plain rules, no model.
On by default for Claude Code: follow and answer a session from claude.ai/code or the Claude app.


11 · Keys, secrets, safety
The third step of the ⚷ Keys page keeps your projects' secrets under a name, encrypted for your Windows user and never shown again. Each one goes only to the sessions you give it to, in its program's environment when it starts: not in a file, not to the others.
DOTENV_PRIVATE_KEY here, give it to the session, dotenvx run -- works as it is.
A level, a file access and git rights for each session, with defaults per group. Agent Manager fills Claude Code's and Codex's own settings; it adds no permission system of its own.
~/.ssh, .env files and your keys: a guard rail on the agent's own tools, not a wall.
On 127.0.0.1 only, behind a token. Nothing leaves your computer unless you ask.
OpenAI, Fly.io, Anthropic: DPAPI on Windows, never shown again, never logged.
Agent CLIs run as you installed them. Agent Manager never reads, copies or relays a subscription token.
Tokens per session, per day and per tool, read from the agents' own logs. No invented dollar amounts.
Compare
Running coding agents side by side is common ground, and others do it well. We name no one and keep no score: this is only where Agent Manager GD chose to go further.
Others
Agent Manager GD
It's a lab
Agent Manager GD is built in loops, by the coding agents it now runs, with its owner testing every round. Alpha, open, honest about what works and what does not yet. Run it on your own machine; don't put it on the internet unless you know what you are doing (TCP/IP, TLS, authentication, infrastructure). The AI notice →
The window, the native terminal (libghostty-vt), views, voice. Windows or Linux.
python3 -m engine, standard library only: sessions, hooks, keys, git, MCP, sandboxes.
A private server (tmux -L agmgd): sessions outlive the app, in WSL on Windows.
Claude Code, Codex, git, ssh, flyctl, Second Brain GD: found, never bundled.
Built in loops
Each loop ends with tests, real screenshots and a journal entry, then the next one starts. Pick a dot on the line, a theme, or search.
Website · Nov 2026
The real app filmed on a private virtual display, cut and narrated; this home page rebuilt around it, one section per feature with new screenshots, and this line of loops that scales.
Loop 48 · Cloud
Sandboxes and Online machines are one page, ☁ Cloud. Every new machine has its own Fly.io app and a door: « Bind to a public URL » beside « Bring it to my localhost » makes one of its ports public over HTTPS, live, without a restart; Unbind closes it.
Loop 47 · Cloud
The Online page holds many services: four tiles for the whole, then one line each with its gate, address, port, size and cost. A click unfolds its door and everything about it.
Loop 46 · Cloud
An agent can ask for a cloud sandbox or an Online machine, drive its sessions, and ask to expose it. What costs money or faces the internet waits in « They wait for you » until you approve it.
Loop 45 · Core
Each expert keeps its lessons in your Second Brain and reads them on the next project. Experts go to cloud machines too, and the orchestrator can hand a task to one, never giving it a key its own session does not hold.
Loop 44 · Core
Agents that know a trade, not a product: a role, skills of the open Agent Skills format (Anthropic's to start, pinned and updated on your OK), the keys they need. Launched in any repository, in Claude Code or Codex alike.
Loop 43 · Cloud
An Online machine takes its repository and its secrets, and its service comes back by itself after a restart (a start script on its volume). Its agent knows the rules: test on your localhost first, then you expose it.
Loop 42 · Safety
Your own gh, read at a glance: who is logged in and where its token is kept. A deploy key added in one click, a new repository private by default and connected at once. The token never passes through Agent Manager, and agents get each repository's own key.
Loop 41 · Cloud
A Fly.io machine for a website or an API: its own address, a volume, and one door, 443 in HTTPS toward its service; the database and the rest stay private on the machine. Private until you expose it, closed again in one click.
Loop 40 · Cloud
A sandbox's card lists the sessions on its machine and opens one more beside them: Claude Code, Codex or a shell, no new machine. Each has its own terminal and connection there; an agent gets its own git worktree, on its own branch.
Loop 39 · Safety
Keys › Secrets keeps your projects' keys under a name, encrypted, never shown again; each goes only to the sessions you give it to, in its environment when it starts. A card per secret with who holds it and fourteen bars of handovers. Works with dotenvx as it is.
Loop 38 · Cloud
Every sandbox made here stays listed, destroyed ones too, with its life over thirty days and what it cost. On a running one, Tunnel and ports: bring a port to your localhost, a SOCKS proxy, a way back you allow, a wire per port with what runs each way.
Loop 37 · Git
One repository per line with its host, its key's state and age, what that key opens and the folder's branch; a line unfolds to the key and the metro map of its branches. Azure DevOps connects with the RSA key it asks for.
Loop 36 · Control room
It reads Windows, WSL and the distribution before the programs and the engine, says why the engine does not answer, installs what is missing on your click in a terminal you see, and has an engine room: memory by session, from the last hour to seven days.
Loop 35 · Control room
Groups and sessions move on screen before the engine has answered; a dictation is written once more, in one piece.
Loop 34 · Voice
What you say appears as you say it, stays in a box, and Enter sends all of it.
Loop 33 · Control room
Measured, then fixed: opening a session again went from 80 ms to 8.
Loop 32 · Control room
The list and « They wait for you » in lines: twenty sessions where eight tiles fitted.
Loops 27–31 · Safety
Level, file access and git rights per session, changed on an open agent on its own conversation; one view for all of them.
Loops 25–26 · Brain & agents
One agent plans across your repositories, asks first, delegates in worktrees with the tool and model your rules choose, and hands you each result.
Loop 24 · Safety
From Ask before acting to No permission; paths no agent may touch at any level.
Loop 23 · Control room
Approve and Refuse answer at once; one notification per request.
Loop 22 · Brain & agents
A live badge, a ✦ light per session, a pulse when an agent uses the Brain.
Loops 19–21 · Website
Twelve guides with a real screenshot per step, the Docs, one command that checks it all.
Loops 17–18 · Website
Every screenshot from the real app, on a fictional workspace, in one command.
Loop 16e · Brain & agents
The agent orients in the Brain, then proposes your words, decisions and its journal.
Loops 12–16 · Cloud
Fly.io machines per agent, tunnels, budgets that only pause; usage per session and day; the installer.
Loops 10–11 · Voice
Encrypted keys, F8 dictation, F9 commands.
Loops 8–9 · Git
Diff review with line comments; deploy keys made for you.
Loop 7 · Brain & agents
Wrap up, summaries proposed to the Brain, Brief from the last one.
Loop 6 · Brain & agents
A Second Brain key per session; agents pilot agents through MCP.
Loop 5 · Control room
A history that never forgets a conversation.
Loop 4 · Control room
The tree, « They wait for you », Approve from the list, alerts.
Loop 3 · Control room
A local engine and a private tmux server: close the app, the agents keep working.
Loop 2 · Control room
libghostty-vt in Godot, on Linux and native Windows: accents, true colour, mouse.
Your agent lab.
Windows (with WSL) or Linux. Unzip, run the installer for your user, and the Setup check tells you what is missing.
Open Source · MIT Licensed