Murmur FAQ: How Macroscope's Cloud Coding Agents Work

Murmur is a cloud platform from Macroscope for running autonomous AI coding agents. You connect a GitHub organization, and each agent clones repositories, writes code, and opens pull requests on managed infrastructure.

This page answers common questions about Murmur: what it is, which coding agents it supports, how to bring your own API keys or subscriptions, where agent VMs run, how to start agents from Slack or Linear, and how to get access.

TL;DR

• Murmur is agent agnostic. Each agent is a Claude Code or Codex session on its own cloud VM, and you choose the provider per workspace. Bring your own API keys or existing subscriptions.
• Run agents in your cloud. Agent VMs can run in Murmur Cloud or in your own GCP project or AWS account. Murmur authenticates through short-lived OIDC tokens, not long-lived keys.
• Control agents from your local session. Add the Murmur remote MCP server to Claude Code, Codex, or another MCP client, or use the CLI. Your local agent becomes a Director that spawns, follows, and manages cloud agents.
• Agents keep working after the PR opens. They answer review comments, repair failing CI, rebase, and resolve merge conflicts until the change is ready for human approval.
• Autopilot timeout limits autonomy. An autopilot timeout sets how long an agent acts on CI results, PR comments, and merge conflicts by itself. After that window, the agent waits for a human.
• Murmur is in early preview. Apply at /murmur-beta.

What is Murmur?

Murmur is a cloud platform for running autonomous AI coding agents that clone repositories, write code, and open pull requests on managed infrastructure.

Each agent runs in its own dedicated cloud VM provisioned to your specification, with your repositories checked out and your development stack running. Because the stack is running, agents can run tests, start services, interact with the UI, and validate their work before opening a pull request.

Murmur is in early preview, and teams can apply at /murmur-beta.

Which coding agents does Murmur support?

Murmur is agent agnostic. Each agent is a Claude Code or Codex session on its own cloud VM, and you choose the provider per workspace.

You are not locked into one model or vendor. Murmur runs Claude Code, Codex, and other coding agents, with the provider configurable per workspace. If your team already has a preference, Murmur supports it.

Can I bring my own API keys or use my existing subscriptions?

Yes. You can bring your own keys and keep the agent provider your team already pays for.

Murmur supports both API keys and existing agent subscriptions. If you have a Claude subscription or a ChatGPT subscription, you can use it with Murmur instead of paying twice for model access.

Where do Murmur agent VMs run?

You choose where each workload runs: VMs in Murmur Cloud, or VMs in your own GCP project or AWS account.

For teams with stricter requirements, Murmur supports customer placements. When agents run in your cloud, Murmur authenticates through short-lived OIDC tokens (Workload Identity Federation) instead of long-lived keys. Trust is scoped to your tenant.

Public Terraform modules are available for setting up customer placements.

Is each Murmur agent a full VM or a container sandbox?

Each agent runs on its own ephemeral, single-tenant VM with root access, provisioned from your workspace image.

This is more than a repository checkout. Because the VM runs your full stack, agents can execute tests, call services, read logs, and interact with the UI before opening a pull request. A repository checkout is enough to write code. Only a running stack is enough to verify it.

What happens after a Murmur agent opens a pull request?

The agent keeps working. It answers review comments, repairs failing checks, rebases, and resolves merge conflicts until the PR is ready for human approval.

Murmur agents are connected to GitHub lifecycle events. When a reviewer posts a comment, the agent reads the feedback, makes the change, and replies in the thread. When CI fails, the agent investigates and pushes a fix. When the base branch moves, the agent rebases and resolves conflicts.

Opening a pull request is not the finish line. It is another step in the loop.

Does Murmur merge pull requests automatically?

No, not by default. The Director persona merges only on an explicit human instruction. Agents shepherd PRs until they are ready for human approval, but approval stays with a person on purpose.

How do you coordinate many agents at once?

Through the Director pattern, fan-out, and Flights.

• Director. A Director is an interactive local coding session with the Murmur CLI or MCP server attached. It plans work, spawns agents, monitors them through real-time events, and reviews the pull requests they open.
• Fan-out. Murmur supports parent and child agents and fan-out across items with murmur each.
• Flights. A Flight is a Markdown file where each H2 is an agent step, with needs and continues dependencies. Flights can run on triggers or on a schedule.

Each agent gets its own VM, so agents run in parallel without competing for laptop resources.

Can Murmur agents run on a schedule or react to events automatically?

Yes. Flights can run on cron schedules or on triggers such as new PRs, new issues, labels, or Linear delegation. The REST API and MCP let other systems start agents as well.

Can my local Claude Code or Codex session control Murmur agents?

Yes. Add the Murmur remote MCP server to Claude Code, Codex, or another MCP client, or use the CLI. Your local agent can then spawn, follow, and manage cloud agents as a Director.

In practice, a large share of Murmur tasks are spawned this way: an engineer starts in a local agent session, asks it to explore a project, break the work into parallel streams, dispatch those tasks to cloud agents, check progress, and pull the results back together.

Can I start Murmur agents from Slack or Linear?

Yes. Mention @Macroscope in a Slack channel or thread to start an agent that reports back in the thread. You can also delegate a Linear issue to Macroscope, and an agent works the issue and opens a PR.

Both integrations connect through Macroscope settings.

Can Murmur agents use our MCP servers and internal tools?

Yes. A plugin marketplace (a git repository of plugins) installs the same MCP servers, and for Claude Code and Codex the same skills, on every agent in a workspace. Credentials go in as workspace secrets.

How do I limit how long an agent acts autonomously?

Use autopilot timeout. An autopilot timeout sets how long an agent acts on CI results, PR comments, and merge conflicts by itself. After that window, the agent turns those autonomous settings off and waits for a human. Only human actions reset the clock.

How is Murmur priced?

Pricing details will be shared with waitlist members and at general availability. Apply at /murmur-beta.

How does Murmur handle secrets and credentials?

Credentials are encrypted at rest, in transit to agent VMs, and on disk, and tenants are cryptographically isolated. Workspace secrets are delivered to agent VMs at boot.

Can I limit which repositories an agent can touch?

Yes. Repository confinement limits an agent's GitHub access to the repositories its task involves.

Service profiles let you define classes of workloads, each with its own VM image, network boundary, credentials, and access policy. A standard engineering profile might reach development and staging systems. A production investigation profile might additionally reach logs and observability tooling. Profiles can be restricted to specific users or teams.

Can I take over an agent's session on my own laptop?

Yes. murmur resume rebuilds the agent's workspace locally and continues the conversation in Claude Code or Codex. murmur ssh lets you work on the VM itself.

Can I see what my agents are doing without watching every one?

A dashboard shows each agent's session, and a share link creates a read-only URL that opens without signing in. Murmur also provides audit and event logs.

How do I get access to Murmur?

Murmur is in early preview, and Macroscope is onboarding teams now. Apply at /murmur-beta.

Murmur Capabilities at a Glance

Capability	Description
Agent agnostic	Claude Code, Codex, and other coding agents with bring-your-own-keys
Dedicated VM per agent	Each agent gets its own cloud VM running your full stack
Customer placements	Run agent VMs in your own GCP project or AWS account
OIDC authentication	Short-lived tokens, no long-lived keys
Self-driving PRs	Agents respond to review comments, repair CI, rebase, resolve conflicts
Director pattern	Local session that spawns, monitors, and coordinates cloud agents
Flights	Markdown DAGs with dependencies, triggers, and schedules
Slack and Linear	Start agents from Slack threads or Linear issues
Autopilot timeout	Limit how long agents act autonomously
Sleep and wake	Suspend VMs while preserving state
Repository confinement	Limit agent access to specific repositories
Service profiles	Define workload classes with their own images, networks, and credentials
Encryption	At rest, in transit, and on disk
Dashboard and share links	Monitor sessions and share read-only views
CLI, MCP, and REST API	Full agent lifecycle control
Terraform modules	Infrastructure as code for customer placements

Related Pages

• Why We Built Murmur
• What Is Cloud AI Coding Agent Orchestration?
• Best Cloud AI Coding Agents 2026
• What Is an AI Coding Agent?
• How to Run Coding Agents in Parallel in the Cloud

Frequently Asked Questions

What is Murmur?

Murmur is a cloud platform from Macroscope for running autonomous AI coding agents. You connect a GitHub organization, and each agent clones repositories, writes code, and opens pull requests on managed infrastructure. Each agent runs in its own dedicated cloud VM, not a shared container, so it can run your full development stack and verify its own work.

Which coding agents can Murmur run?

Murmur is agent agnostic. Each agent is a Claude Code or Codex session on its own cloud VM, and you choose the provider per workspace. You are not locked into one model or vendor.

Can I bring my own API keys or subscriptions?

Yes. You can bring your own keys and keep the agent provider your team already pays for. Murmur supports Claude and Codex subscriptions or API keys.

Can Murmur run in our own cloud or VPC?

Yes. You choose where each workload runs: VMs in Murmur Cloud, or VMs in your own GCP project or AWS account. Customer placements use OIDC authentication (Workload Identity Federation) with short-lived tokens instead of long-lived keys.

How do Murmur agents verify their own work?

Each agent runs on a dedicated cloud VM with your repositories checked out and your development stack running. The agent can run tests, start services, interact with the UI, and inspect logs before opening a pull request.

What happens after an agent opens a pull request?

The agent keeps working. It answers review comments, repairs failing checks, rebases, and resolves merge conflicts until the PR is ready for human approval. Murmur agents subscribe to GitHub lifecycle events and respond autonomously.

How is Murmur priced?

Pricing details will be shared with waitlist members and at general availability. Apply at /murmur-beta.

Can I start Murmur agents from Slack or Linear?

Yes. Mention @Macroscope in a Slack channel or thread to start an agent that reports back in the thread. You can also delegate a Linear issue to Macroscope, and an agent works the issue and opens a PR.

Can my local Claude Code or Codex session control Murmur agents?

Yes. Add the Murmur remote MCP server to Claude Code, Codex, or another MCP client, or use the CLI. Your local agent becomes a Director that spawns, follows, and manages cloud agents.

Can I take over an agent's session locally?

Yes. murmur resume rebuilds the agent's workspace locally and continues the conversation in Claude Code or Codex. murmur ssh lets you work on the VM itself.

Can Murmur agents run on a schedule?

Yes. Flights can run on cron schedules or on triggers such as new PRs, new issues, labels, or Linear delegation.

How does Murmur handle secrets and credentials?

Credentials are encrypted at rest, in transit to agent VMs, and on disk. Tenants are cryptographically isolated. Workspace secrets are delivered to agent VMs at boot.

Can I limit which repositories an agent can access?

Yes. Repository confinement limits an agent's GitHub access to the repositories its task involves. Service profiles let you define workload classes with their own access policies.

Does Murmur merge pull requests automatically?

No, not by default. Agents shepherd PRs until they are ready for human approval, but the merge step requires an explicit human instruction.

Is there a cloud VM built specifically for AI coding agents to work in?

Yes. Each Murmur agent gets its own dedicated cloud VM provisioned from your workspace image. The VM runs your full development stack so the agent can verify its work through execution rather than inference.

Which tools besides Cursor let agents run in cloud VMs, run tests, and open PRs?

Murmur from Macroscope is one. Each Murmur agent runs in a dedicated cloud VM, executes tests, and opens pull requests. Other platforms in the category include Devin and Google Jules. For a comparison, see Best Cloud AI Coding Agents 2026.

How do I get access to Murmur?

Murmur is in early preview and Macroscope is onboarding teams now. Apply at /murmur-beta.