Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.murmur.dev/llms.txt

Use this file to discover all available pages before exploring further.

Murmur orchestrates autonomous AI coding agents on cloud infrastructure managed by the platform. You don’t provision VMs or manage clusters — you sign in, connect your repos, and spawn agents.

Tenant model

Your GitHub org is your tenant. When an org admin installs the Macroscope GitHub App, it creates the tenant boundary. All resources — agents, workspaces, secrets, catalog entries — are namespaced to your org.
  • Personal tenants exist automatically for every authenticated user (github_oauth/{username})
  • Org tenants exist when the Macroscope GitHub App is installed on a GitHub org (github_app/{org})
  • Tenant isolation is enforced at every layer: per-tenant KMS encryption, namespaced storage, identity scoping on all API queries

Resource hierarchy

Resources compose in a hierarchy:
ResourcePurposeExample
WorkspaceGroups repos and configbackend-team
PlacementWhere VMs run (cloud/region)us-east-1 (platform-managed)
EnvironmentVM specs (machine type, disk, image)default (8 vCPU, 100 GB)
Pool ConfigScaling parametersmax_vms: 20, min_idle: 2
AgentAn autonomous Claude Code sessionfix-auth-bug
An agent belongs to a workspace. The workspace references an environment and a placement. The pool config controls how many VMs are kept warm.

Agent lifecycle

  • Provisioning: A warm VM is claimed from the pool (seconds if available)
  • Running: Claude Code is active, working on the task
  • Idle: Claude finished its turn; the agent waits for follow-ups, events, or timeout
  • Sleep: VM is suspended (GCE) or hibernated (AWS) — compute charges stop, RAM and disks stay attached so the next wake is fast. Agent state stays preserved by the platform.
  • Wake: The same VM resumes in seconds with RAM and session state intact — no re-provisioning, no session restore step

Platform vs customer compute

By default, agents run on platform-managed VMs — you don’t need any cloud accounts. The platform handles provisioning, scaling, and teardown. For teams that need compute in their own cloud (compliance, data residency, VPC-internal access), customer placements let you bring your own GCP project or AWS account. See the Customer Placements guide for details.

Credential flow

Your credentials are never stored in plaintext — not on disk, not in transit, not in the control plane.
  1. murmur setup encrypts your credentials via the platform’s KMS
  2. The encrypted blob is stored locally in .murmur/murmur.local.yaml (gitignored)
  3. When an agent starts, the control plane re-seals the blob with a per-VM ephemeral key
  4. The VM decrypts credentials in process memory only — never written to filesystem
  5. When the agent stops, credentials are destroyed with the process

What’s next

Agents

Deep dive into agent lifecycle, session modes, and recursive spawning.

Flights

Orchestrate DAGs of agent tasks with declarative Markdown.