Preface¶

On July 30, 2026, Conductor released version 0.78.0 and officially launched Conductor Cloud. CEO Charlie Holtz and co-founder Jackson de Campos have evolved this coding Agent management tool from a Mac application relying on local Git worktrees into a shared cloud workspace: Agents run continuously in isolated microVMs, so tasks will not be interrupted when developers close their laptops; team members can collaborate in real time via workspace links; and external systems can programmatically schedule Agent sessions through the Conductor API.

Conductor is positioned as a “multi-Agent parallel IDE” — it can run coding Agents such as Claude Code, Codex, Cursor, and OpenCode in parallel on different branches of the same repository, with independent workspaces, terminals, Diff, and Review processes for each. The launch of the Cloud version solves two pain points of the local worktree solution: execution environment is bound to the local machine and asynchronous long tasks cannot be resumed across devices. This article sorts out the core capabilities and usage methods of the Cloud version based on the Conductor official Changelog, API documentation and public reports.

What is Conductor Cloud?¶

Conductor Cloud migrates regular Conductor workspaces to run on independent cloud machines. The interaction flow is consistent with the local version: you can chat, view Diffs, use the Git panel, and open terminals in the Conductor desktop client. The changes are mainly reflected in two points:
1. Persistent execution: When you close the app or close your laptop, the Agent will continue to work in the cloud sandbox; when you reopen the workspace later, you can continue the conversation from the breakpoint.
2. Reduce local load: The worktree, development processes, and Agent processes of each workspace run on the cloud microVM instead of occupying local Mac resources.

According to official instructions, Cloud workspaces run in isolated microVMs, with pre-installed repositories and dependencies, providing Agents with a persistent execution environment. In terms of underlying infrastructure, Conductor previously disclosed in the Vercel official blog that Cloud Workspaces are built on Vercel Sandboxes; Holtz said that users “can barely perceive the difference between local and cloud environments”.

Conductor adopts a model/Agent neutral strategy: it does not bind to a single LLM vendor, but connects to the user’s existing subscriptions or API Keys of Claude Code, Codex, Cursor, etc., and Conductor itself acts as the orchestration and control layer. The capability limits and usage policies of Agents are still determined by their respective vendors.

Multi-person collaboration: from a single-user Mac tool to a team product¶

On July 23, Conductor launched the Alpha test of Cloud multi-person collaboration and API for Pro plan users; on July 30, the official version further opened Cloud organization onboarding, shared repository settings, and workspace management.

The core collaboration capabilities of the Cloud version include:
- Workspace link sharing (shortcut ⌘⇧C): Team members can directly enter the same workspace without PR links.
- Real-time online status: View who is active in which workspace and follow tasks of interest.
- Collaborative prompts: Multiple people can issue commands and monitor progress together in the same Agent session.

Holtz described it on X as an alternative to “heavy worktree Agent workflows” — previously developers needed to maintain multiple repository copies locally, and Cloud集中隔离环境到远端, and the collaboration interface is unified in the Conductor client.

Regarding the mobile terminal, Holtz said that the iPhone application was approved by Apple TestFlight on July 30 and invitations have been sent out; the official website still marks iOS as “coming soon”. The first TestFlight version only supports Cloud workspaces — which is consistent with the product architecture: the persistent remote sandbox can supervise Agent progress from a mobile phone without a Mac being online.

Conductor API: Programmatically schedule Agents¶

Launched simultaneously with the Cloud version is the Conductor API (Beta), with the endpoint located at https://api.conductor.build/v0, and the complete contract can be found in the OpenAPI specification.

The core operations supported by the API include:
- Create Cloud workspaces, start Agent sessions, send prompts, and read replies.
- Archive workspaces, query session status and full transcripts.
- Read-only retrieval of organization-wide session transcripts via POST /v0/sql for building internal dashboards or Bots.

The typical calling process is as follows.

1. Create an API Key at app.conductor.build/users/api-keys and list available projects:

curl https://api.conductor.build/v0/projects \
  -H "Authorization: Bearer YOUR_API_KEY"

2. Create a Cloud workspace and start the first Agent session:

curl -X POST https://api.conductor.build/v0/workspaces \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "projectId": "PROJECT_ID",
    "name": "my-workspace",
    "agent": "codex",
    "model": "gpt-5.5"
  }'

3. Send a prompt to the session and poll the status:

curl -X POST https://api.conductor.build/v0/sessions/SESSION_ID/messages \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"content": "Fix the failing unit test in src/utils.ts"}'

curl https://api.conductor.build/v0/sessions/SESSION_ID/status \
  -H "Authorization: Bearer YOUR_API_KEY"

The official documentation emphasizes that the agent parameter must match the model (supports claude, codex, cursor, etc.); after the prompt is enqueued, the session status will first show idle, and it will not be considered complete until it changes to working and then back to idle. Inside the Cloud workspace, you can also use the environment variables CONDUCTOR_API_URL and CONDUCTOR_API_KEY to let the Agent itself call the API, realizing nested scheduling of “Agent managing a fleet of Agents” — the official Cookbook provides prompt templates for Plan/Implement, multiple PR parallelization, Daily Report, etc.

Existing users have built internal tools such as Slack Bots and custom mobile clients based on the API.

Competitive landscape with Cursor Cloud VM and Sculptor¶

In 2026, the “Agent infrastructure layer” has become a new battlefield for developer tools. Conductor Cloud indirectly competes with the following products:

The differentiation of Conductor lies in: a neutral orchestration layer across Agent vendors. Holtz’s judgment is that coding work is becoming asynchronous and scattered across multiple machines, multiple team members, and automated systems; developers need a control plane that does not bind to a single model, and Cloud workspaces + API extend it from a single-user Mac application to team budgets and external system integration — Agent tasks can be assigned by another application or Bot via the API before the developer opens Conductor.

Regarding the company background, Holtz and de Campos met at Brown University, entered Y Combinator in the summer of 2024 with the open-source AI code editor Melty, and later focused on Conductor. In March 2026, Conductor completed a $22 million Series A round (led by Spark Capital, Matrix Partners, etc.), with total financing of approximately $24 million. The company claims to have grown about 10 times since January 2026, and says more than 100,000 “builders” use its products (all data disclosed by the company itself, not third-party audited).

Pricing and access methods¶

Conductor Cloud is a paid plan feature:
- Pro: $50/month, including Cloud workspaces and API access.
- Teams: $60 per user per month, adding centralized billing, admin portal, and SSO; team sales are currently by invitation only.
- Free: Still limited to parallel local workspaces on Mac.

Access steps: Register for Pro → Create a Cloud organization → Create a workspace in the desktop client or via the API. Version 0.78.0 also improved onboarding processes such as Cloud compute environment setup guides, environment variable imports, and multi-GitHub organization connections.

Please note that Cloud workspaces still have some Beta limitations — for example, Codex’s Goals function is temporarily unavailable on Cloud, and the API request/response format may change. It is recommended to check the Cloud Beta documentation before use.

Summary¶

Conductor Cloud migrates the execution environment of coding Agents from local worktrees to persistent cloud microVMs, filling the gaps of “resume after closing the laptop” and “multi-person collaboration”; the Conductor API exposes the same set of capabilities to external systems such as Slack Bots, CI pipelines, or OpenClaw, promoting coding Agent development from interactive conversations to asynchronous, orchestratable engineering workflows.

For developers who heavily use Claude Code / Codex / Cursor on a daily basis, if you are already accustomed to Conductor’s parallel branch workflow, the Cloud version is worth trying under the Pro plan — especially for scenarios that require long-term background tasks, cross-timezone collaboration, or integrating Agents into existing toolchains via the API. In terms of competition, Cursor, Sculptor, and others are also competing for the “Agent control plane” position; Conductor’s bet is: models will change their “vibes” every few months, but a neutral orchestration layer can exist for the long term.