Horme
Private beta, autumn 2026
AI agent orchestration for multi-agent coding
Horme coordinates Claude Code and Codex across connected work, keeps parallel and dependent tasks organized, and shows you what needs your attention. We are building Horme toward a prompt first control layer, where you describe the mission and Horme routes the work.
One system. Multiple coding agents.
you are on the list.
we will write to from hello@usehorme.com.
preview mode: WAITLIST_ENDPOINT is empty, so this address was not stored.
claude code, 
codexmission fix mobile UI, investigate slow deletesillustrative, product direction
youFix the mobile UI, investigate why deletion is slow, and have Codex review the finished work.
hormeSplitting this into steps and routing them to your agents.
claude code
Fix mobile UI
Running
claude code
Investigate slow deletes
Running
codex
Review the result
Waiting
2 working · 1 waiting · 0 needs you
§1The problem
Opening one coding agent is easy. Running several turns you into their coordinator: what is running, what finished, what depends on what, and who needs you. Stop babysitting your agents, one terminal at a time.
i.
Claude is working in one place. Codex is somewhere else. Another task finished ten minutes ago. The more agents you run, the more places you have to check just to understand what is happening.
ii.
You decide what should run next, remember which work depends on what, copy context around, notice when something stops, and keep the whole plan in your head. Your agents are fast. Their coordination is manual.
iii.
You see an answer, change direction, remember another task, or an agent discovers something new. AI coding is iterative, so a static plan goes stale the moment you drew it. That is why Horme is not a workflow builder.
§2The solution
Horme organizes the work across Claude Code and Codex: dependent work waits, independent work moves in parallel, available in the private beta today. We are building toward simply telling Horme what you want done.
Describe the outcome in normal language, the way you would brief a coding agent, not nodes and edges. This is the direction Horme is being built toward.
conversational AI agent orchestration
Horme organizes work across agents, keeps dependent work waiting for what it needs, and lets independent work move in parallel.
AI agent orchestration
See what is running, waiting, finished, failed, or asking for you, without rebuilding the state from five terminals.
AI agent monitoring
Add another instruction, change direction, switch agents, or ask why something is stuck. This is where Horme is going: today, questions, permissions, stalls and failures already come back to you through Needs You.
human-in-the-loop AI
§3Differentiation
It coordinates them.
Keep using the coding agents you already trust. Claude Code and Codex do the coding. Horme sits above them as the control layer for multi-agent work: one conversation, one execution view, one place to step in.
§4In conversation
An instruction can start a mission, or steer one that is already running.
mission settings page and startup performanceillustrative product direction
youFix the settings page and investigate why startup is slow. Use Claude for the UI.
hormeI am splitting this into two pieces.
claude code › fix settings UI
codex › investigate startup performance
youWhen Claude finishes, have Codex review the UI too.
hormeAdded. The review will wait for Claude's work.
This is not a live transcript. It is the direction Horme's conversational layer is being built toward; today, missions are created and organized through Mission Control.
§5The work graph
A mission can branch into parallel work and come back together later.
workflow auth middleware6 steps, illustrative
claude code
implement API
codex
security review
claude code
write tests
claude code
fix findings
codex
verify
AI agent orchestration means coordinating multiple AI agents so they can work on connected tasks without the human manually managing every transition.
§6How it works
Pair your machine and choose the repository you want to work on. Horme detects the coding agents you already have installed, and Claude Code and Codex keep using their existing accounts.
$ npx horme connect
Describe the job in normal language, the way you would brief a coding agent. This is the direction Horme is being built toward; today, missions are created and organized in Mission Control.
product direction
For larger jobs, Horme will be able to show how it intends to split and route the work before anything runs. This is the workflow Horme is being built toward, not a shipped feature yet.
product direction
Independent work runs in parallel. Dependent work waits for what it needs. Horme starts what is ready and holds what must wait.
running · 2, waiting · 1
Answer an agent's question, approve a permission, or step in on a failure or conflict through Needs You today. Adding a follow-up instruction mid-mission is where Horme is headed next.
needs you · 1
§7Trust
i.
Agents execute locally through the Horme runner, using the accounts and repositories already on your computer. Horme does not clone your repository or store a full copy of it.
ii.
Horme keeps routine work moving and brings you back for the moments that need human judgment, such as questions, permissions, failures and conflicts.
iii.
Runner access tokens are stored only as one way hashes, pairing codes expire after a single use, and privileged database credentials never reach the browser.
§8FAQ
Horme is an AI agent orchestration tool for coding. It coordinates coding agents such as Claude Code and Codex so developers can manage multi-agent work from one place. It is being built toward a conversational model, where you describe the mission and Horme coordinates the work across agents.
AI agent orchestration is coordinating multiple AI agents so the right work runs at the right time, dependent work waits for what it needs, and a human steps in when necessary.
No. Claude Code and Codex still do the coding. Horme coordinates the work around them.
Horme's current beta includes execution graphs and connected work. The product is moving toward a prompt first model, where you describe the mission in normal language and Horme creates the execution structure for you.
Horme runs your coding agents on your own machine, against repositories you choose. Horme is not a cloud execution product.
Claude Code and Codex today.
§9Private beta
Horme is for developers who already use AI coding agents and are starting to hit the coordination problem. If you use Claude Code, Codex, or several coding tasks at once, join the private beta.
you are on the list.
we will write to from hello@usehorme.com.
preview mode: WAITLIST_ENDPOINT is empty, so this address was not stored.
Questions, or want to be an early tester? hello@usehorme.com