Termdock01 / Agent-to-agent
Let agents hand off the work.
Claude Code implements, Codex reviews. Agents can open terminals, send each other tasks and follow the results.
See how it worksAgent workflows that run without you watching
Nine workflows Termdock is built for: fixing things from your phone, agents dispatching and reviewing each other, code graph lookups, scheduled and webhook runs, remote machines, previews and cost.
Connected, not just side by side
Hand off work. Keep the team moving.
02 / Connected machines
Put every machine to work.
Connect your laptop and other computers. Run agents where the project lives and coordinate their terminals from one workspace.
See how it works03 / Phone control
Pick it up from your phone.
Switch terminals, type instructions and respond to your agents in a phone browser. No companion app required.
See how it works04 / Human-in-the-loop
Intervene where it matters.
Bring supported agent requests into Workstream instead of checking every terminal for a blocked prompt.
See how it works05 / Terminal comparison
Termdock, tmux, Orca or Herdr?
Compare agent coordination, persistence, remote control and worktree handling, with the differences spelled out.
See how it worksWhat these workflows rely on
| Feature | Termdock | Without it | What changes |
|---|---|---|---|
| Phone browser access | Wait until you are back | Read and type into the desktop terminal from the phone, no app to install | |
| termdock CLI and Terminal API | You copy output between tabs | An agent starts other agents and follows their output | |
| Agent Workstream | Check every tab for prompts | Permission requests and questions from hooked Claude Code, Codex and Gemini CLI in one panel | |
| Schedules with a precheck | Start runs by hand | Runs start on time and skip when there is nothing to do | |
| Webhook into a tab | Paste the CI log yourself | A script or CI step posts a message that arrives as a prompt | |
| Discord and Telegram notifications | Come back to check | Pinged when a watched session goes idle or an agent calls termdock notify | |
| LAN, SSH and takeover | Walk to the other machine | Drive another machine's session, then give it back | |
| Code graph for agents | Agents grep for callers | Definitions, callers and dependencies in 9 languages | |
| Browser panel | Switch to another browser | The dev server or local HTML file opens next to the agent's terminal | |
| AI usage dashboard | Check each provider's page | Claude Code and Codex cost, filtered by workspace | |
| Terminals, split panes, workspaces | Standard in agent terminals | Included, with a file viewer and browser in the same layout |
Fix a failing job from your phone
Phone browser, Discord or Telegram
Away from the desk
Phone browserDiscordTelegram
Best For:
You left a long job or an agent running on the desktop, and something breaks while you are out.
KEY FEATURES FOR THIS USE CASE
Phone browser accessNotificationsDiscord remote controlTailscale / ZeroTier
You get pinged
- •You /watch a long command in a background tab; when its output goes idle, Termdock pushes it to Discord or Telegram
- •An agent that is blocked can push its own message with termdock notify
- •A session's first push opens its Discord thread, where you can ask for a screenshot of its screen
You open the terminal in the phone browser
- •Turn on browser access in Peer Sessions on the desktop and pair the phone once; no app, domain or certificate
- •You see every local workspace and land on the terminal open on the desktop; a reload brings you back to the last terminal
- •On the same network it connects directly; away from home, use a Tailscale or ZeroTier address
What you do
- •Read the error on the same screen the desktop shows
- •Type the fix or the next command; for a yes or no, a reply in the Discord thread is enough
- •Revoke any paired browser from the desktop when you no longer need it
It broke while you were out, and you fixed it from the phone
One agent hands work to other agents and follows them
Claude Code dispatching Codex and Antigravity
Agent-to-agent dispatch
Claude CodeCodexAntigravity
Best For:
Developers who already let one agent plan and want it to hand parts of the job to other agents.
KEY FEATURES FOR THIS USE CASE
termdock CLITerminal APIBuilt-in skillsAgent Workstream
Claude Code opens a terminal for Codex
- •Bundled skills, installed into Claude Code and Codex automatically, teach the termdock CLI and Terminal API; the CLI is installed from the app's settings
- •termdock session create --workspace <id> --name codex --background --json starts a headless terminal
- •It starts Codex there and hands over the task with termdock session submit, which can hold the prompt until the session is idle
It follows the work instead of guessing
- •termdock session output <id> --follow streams the other session while it works
- •termdock session status <id> shows whether that terminal is idle or still executing; for agent sessions, the Agent Session HTTP API reports whether a dispatch is done or waiting on permission
- •It reads the final screen and brings the result back into its own work
- •It renames and arranges the terminals so you can follow along
What you do
- •Once Claude Code, Codex or Gemini CLI hooks are set up (termdock hooks setup), answer their permission requests, questions and exit plans in the Workstream panel
- •Approve, reject or reply there without checking each tab
- •Read the Workstream timeline to see what each agent did
Claude Code plans, Codex and Antigravity do the work, you approve
Claude Code writes, Codex reviews
A second agent reads the change before you do
Writer and reviewer agents
Claude CodeCodex
Best For:
Developers who want a second agent to go through every change before they review it themselves.
KEY FEATURES FOR THIS USE CASE
termdock CLIBuilt-in skillsGit diff reviewAgent Workstream
Claude Code writes the change
- •You give the task to Claude Code in its own tab
- •It edits the code and runs the tests there
- •When it is done, it opens a background terminal for the reviewer
Codex reviews in another terminal
- •Claude Code starts Codex in that terminal and submits a review prompt
- •It follows Codex's output with termdock session output --follow until Codex stops
- •It reads the review from Codex's screen and fixes the points it agrees with
What you do
- •Read the final change in Git diff review, with both sides highlighted line by line
- •Answer any permission requests in Workstream
- •Open the Codex tab when you want the full review text
Two agents check the change before you do
Agents check who calls a function before they change it
Code graph through the AST API
Structural context for agents
Claude CodeCodexCode graph
Best For:
Codebases where a changed signature breaks callers the agent never opened.
KEY FEATURES FOR THIS USE CASE
Code graphtermdock-ast skillCallers and dependenciesLSP call lookup
Give the agent the graph
- •The bundled termdock-ast skill lets the agent query the AST API
- •Definitions, callers and dependencies are indexed for 9 languages: JavaScript, TypeScript/TSX, Python, Rust, Go, Java, C, C++ and C#
- •The agent can also ask for the impact of changing a single symbol
It looks before it edits
- •The agent asks for a function's callers before it changes the signature
- •Express, Fastify, Koa, FastAPI and Flask routes are graph nodes, so it can see which route reaches a handler
- •Termdock asks the language server for call relations when the static graph is not enough
What you do
- •Before approving, ask the agent which callers it checked
- •Open the call graph to check a single function yourself
- •Send the change once the affected callers are covered
The agent knows who calls the function it is changing
Agents run overnight on a schedule
Schedule, precheck command, keep-alive
Scheduled agent runs
SchedulesClaude CodeCodex
Best For:
Work you want done by morning, such as dependency bumps, flaky test triage or a nightly report.
KEY FEATURES FOR THIS USE CASE
SchedulesPrecheck commandKeep-alivePrompt library
Set the time, the prompt and the precheck
- •Pick a cron expression or a one-time run; at that time Termdock creates a terminal and sends the prompt
- •Pick the prompt from your Prompt library
- •Add a precheck command, such as "new commits since yesterday"; when it does not match, the run is skipped and the reason recorded
Keep a long session moving
- •Keep-alive sends a prompt when the terminal sits idle past your threshold, and can use a Prompt library entry
- •A workflow runs in the active terminal and sends its prompts one after another to the agent in the foreground
- •If you /watch the session, Discord or Telegram tells you when its output goes idle
What you do in the morning
- •Read the run history: what ran, what was skipped and why
- •Open the terminals the runs left behind and check the results
- •Check the night's Claude Code and Codex cost in the AI usage dashboard
Set it in the evening, read the results in the morning
A script or CI step pushes a prompt into a tab
Webhook URL into a chosen terminal tab
Webhook to agent
WebhookCIScripts
Best For:
Anyone with a script or CI job that should hand an event to an agent waiting in a tab.
KEY FEATURES FOR THIS USE CASE
Webhook URLNotificationsDiscord thread
Point the event at a tab
- •Create a Webhook URL for a specific tab in a specific workspace
- •The URL is shown once at creation; revoke it in Settings when you stop using it
- •Your script or CI step POSTs {message, submit} to it; the service listens on 127.0.0.1, so the CI job runs on the same machine or reaches it through your own tunnel
The agent picks it up
- •A failed CI job posts its message to the Claude Code tab for that repo, where it arrives as a prompt
- •The message is written only while the agent is waiting for input
- •If the tab is at a shell prompt with no agent running, the request is refused
What you do
- •Get a Discord or Telegram message if you /watch that tab, or when the agent calls termdock notify
- •Reply in that session's Discord thread to talk to it
- •Review the fix when you are back at the desk
CI fails, an agent is already on it
Run agents on the stronger machine, drive them from the laptop
LAN machines, SSH workspaces, take over and give back
Remote machines and takeover
LANSSH (Beta)Take over
Best For:
People whose desktop or server has the cores and memory, but who spend part of the day on a laptop.
KEY FEATURES FOR THIS USE CASE
LAN pairingSSH workspaces (Beta)Take over and give back
Mount the other machine
- •Machines on the same network find each other; pair once, end-to-end encrypted, no cloud relay
- •Mount the other machine's workspace; it reconnects after a drop or sleep
- •Or add a directory on an SSH host with tmux installed as an SSH workspace (Beta); short disconnects reconnect on their own
Run the agents there
- •Open a terminal in the mounted workspace and start Claude Code or Codex; it runs on that machine
- •Browse the mounted workspace's files and Git status, read-only
- •Close the laptop and the session keeps running per your retention setting; reattach it later
What you do
- •At the desktop, take over the terminal the laptop opened
- •Only one side can type at a time
- •Give it back to the laptop when you leave
Agents on the desktop, you on the laptop
See what the agent built next to its terminal
Browser panel and local HTML preview
Preview agent output
Browser panelClaude CodeCodex
Best For:
Front-end and docs work where you need to see the page the agent just changed.
KEY FEATURES FOR THIS USE CASE
Browser panelLocal HTML previewSession portsSplit panes
The agent starts the page
- •It starts the dev server in a background terminal
- •termdock session ports <id> --json reports the port that terminal is listening on
- •Or it writes a static HTML file
Open it beside the terminal
- •Open a browser tab next to the terminal, or in its own split pane
- •Point it at the localhost address; plain HTTP works for local dev servers
- •Local HTML files preview directly
What you do
- •Check the page while the agent keeps working in the terminal beside it
- •Type what is wrong straight into the agent's terminal
The agent builds it, you look at it in the next pane
Know which project your agents spend on
AI usage dashboard for Claude Code and Codex
Agent cost tracking
Claude CodeCodexUsage dashboard
Best For:
People running scheduled and dispatched agents across several projects who want to know where the quota goes.
KEY FEATURES FOR THIS USE CASE
AI usage dashboardWorkspace filterSchedules
Open the dashboard
- •Claude Code and Codex quota usage and estimated cost in one place
- •24-hour and 14-day charts, plus a yearly heatmap
- •Only Claude Code and Codex are tracked
Filter by workspace
- •Pick a workspace to see what that project's agents used
- •Line up the hours your schedules run with the 24-hour chart
What you do
- •Move or drop schedules that cost more than they return
- •Add a precheck command so runs with nothing to do are skipped
See which project the agents are spending on
Let your agents run the terminal while you are away
Why people running agents choose Termdock
Agents run the terminal
- • termdock CLI and Terminal API: create, read, stream, send input, arrange panes
- • An agent opens another agent's terminal and follows its output
- • Bundled skills are installed into Claude Code and Codex automatically; the termdock CLI is installed from the app's settings
- • AgentSession callbacks let outside monitors send structured events
Work starts without you
- • Schedules: cron or one-time, with a precheck and a trigger log
- • Webhook URL sends a message into a chosen tab
- • Workflow sends prompts one after another to the agent in the active terminal
- • Keep-alive nudges an idle session with a prompt
You hear about it when it matters
- • Discord or Telegram for sessions you /watch, or when an agent calls termdock notify
- • A session's first push opens its Discord thread; type there to talk to it
- • Phone browser access with no app; revoke any paired browser
Agents understand the code
- • Code graph: definitions, callers and dependencies in 9 languages
- • Express, Fastify, Koa, FastAPI and Flask routes are nodes in the graph
- • Agents query it through the bundled termdock-ast skill
You see what agents cost
- • Claude Code and Codex quota and cost in one dashboard
- • 24-hour and 14-day charts plus a yearly heatmap
- • Filter by workspace
Your machines, your network
- • LAN connections are end-to-end encrypted and skip the cloud
- • Public web pages in the browser panel cannot send background requests to localhost, 127.0.0.1 or LAN IP addresses
- • Free for macOS and Windows
Frequently Asked Questions
Which agents does Termdock work with?
Termdock has built-in session support for Claude Code, Codex, Antigravity (agy, which replaced the Gemini CLI integration), GitLab Duo and OpenCode. Any other CLI agent runs in a Termdock terminal too.
How does an agent control Termdock?
Through the termdock CLI and the Terminal API. An agent can create terminals, read and stream their output, send input, rename them and arrange panes, and it can open another agent's terminal and follow its output. Bundled skills are installed into Claude Code and Codex automatically; the termdock CLI is installed from the app's settings.
What happens when I leave my desk?
Agents keep running. Push to Discord or Telegram is off by default. Once you turn it on, sessions you /watch notify you when their output goes idle, and agents can push a message themselves with termdock notify. A session's first push opens a Discord thread for it, and typing in that thread talks to the session. Claude Code and Codex permission requests stay in the Workstream panel on the desktop. You can also open Termdock in your phone browser after pairing it once.
Do my terminals go through a cloud service?
Connections between Termdock machines on your LAN are end-to-end encrypted and do not go through a cloud. Phone browser access works on the same network, or through a Tailscale or ZeroTier address when you are away. Notifications are delivered through Discord or Telegram, so they pass through those services.
Share These Use Cases with Your Team
Send this page to whoever on your team runs Claude Code, Codex or Gemini CLI and still checks every tab by hand.
Give your agents a terminal they can run
Claude Code, Codex, Gemini CLI and other CLI agents drive Termdock through its CLI and API, and they can ping you on Discord or Telegram when they need a decision.
macOS • Windows • Free