> ## 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.

# Pricing

> Murmur is billed with usage-based credits, based on the costs your agents consume, so you only pay for what you use. 

<Callout icon="pencil">
  Last updated: September 23, 2026.
</Callout>

Murmur usage is billed in **Murmur credits** under [usage-based pricing](https://docs.macroscope.com/pricing). One credit = \$0.01. Credits are pre-purchased and burn down as agents consume the components below:

<table>
  <colgroup>
    <col width="146" />

    <col width="190" />

    <col width="199" />

    <col width="175" />
  </colgroup>

  <thead>
    <tr>
      <th>Component</th>
      <th>What it charges for</th>
      <th>How it's billed</th>
      <th>Rate</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>[**CPU Compute**](#cpu-compute)</td>
      <td>The agent VM's machine shape</td>
      <td>Per VM-second, while the VM is running</td>
      <td>Varies by machine type — [see rates](#cpu-compute)</td>
    </tr>

    <tr>
      <td>[**NAT IP**](#nat-ip)</td>
      <td>The VM's public IPv4 address</td>
      <td>Per VM-second, while the VM is running</td>
      <td>\$0.0055 / hour</td>
    </tr>

    <tr>
      <td>[**Storage**](#storage)</td>
      <td>[Boot disk](#boot-disk), [sleep memory images](#sleep-memory-image), and [baked images](#baked-images)</td>
      <td>Boot disk & memory image: per VM-second while those bytes exist. Baked image: hourly.</td>
      <td>Varies by disk type — [see rates](#disk-type-rates); baked images \$0.000077 / GiB-hour</td>
    </tr>

    <tr>
      <td>[**LLM Inference**](#llm-inference)</td>
      <td>Tokens served on a Murmur-provided model credential</td>
      <td>Per model call</td>
      <td>[Published per-token rates](https://docs.macroscope.com/model-pricing#pricing) plus 10% markup</td>
    </tr>

    <tr>
      <td>[**Historical Session Storage**](#historical-session-storage)</td>
      <td>An agent's retained session transcript</td>
      <td>Daily, until the agent is deleted</td>
      <td>\$0.00022 / decimal MB-month</td>
    </tr>
  </tbody>
</table>

On a [platform placement](/concepts/placement), the rates above and in the [appendix](#appendix) are the final prices Murmur charges: multiply one by your usage and that is the amount billed, with nothing further added at billing time. LLM Inference is the one exception, because its per-model rates are not published here — that section states separately what is added to the provider's rates.

Rates are quoted in USD, so multiply a dollar rate by 100 to get credits: the \$0.0055 per hour NAT IP rate is 0.55 credits per hour.

Your billing dashboard in **Settings → Usage → Billing → Murmur Usage Detail** breaks down the cost components for each of your workspaces, owners, and tags, by billing period.

See the [appendix](#appendix) below for more details.

<Note>
  These rates apply to VMs Murmur runs for you. On a [customer-managed placement](/guides/customer-placements), your own cloud account is invoiced for the VMs, disks, and images it runs, and Murmur charges one-eleventh of the listed rates — about 9.09% — for them. Your billing dashboard shows only the costs that apply to your workspace.
</Note>

<Note>
  [`murmur get machine-type`](/cli/get) and [`murmur get disk-type`](/cli/get) report the rate stored on the [catalog](/catalog/machine-type) entry, which is the internal figure Murmur prices from rather than the billed rate. The rates on this page are the ones you are charged.
</Note>

# Appendix

<Info>
  A [pool](/concepts/pools) VM accrues the VM-side components (CPU Compute, NAT IP, and the Storage parts) at the same rates from the moment it is created, before any agent claims it.
</Info>

## CPU Compute

The CPU Compute rate is the [machine type](/catalog/machine-type)'s hourly rate. Costs accrue per VM-second while the VM is running, and stop when the VM stops or sleeps.

Each rate below is the CPU Compute rate for one running VM-hour, in every region the built-in prices — `us-central1`, `us-east1`, and `us-west1` on GCP, `us-east-1` and `us-west-2` on AWS. A built-in carries the same rate in each of them. [**NAT IP**](#nat-ip) and [**Storage**](#storage) are separate components on top.

<Callout icon="face-sleeping">
  **Sleep** indicates whether the machine preserves RAM when an agent sleeps:

  * **Warm** resumes with memory intact and can accrue sleep-memory storage.
  * **Cold** stops without a memory image.
</Callout>

### GCP machine types

<table>
  <colgroup>
    <col width="203" />

    <col width="148" />

    <col width="109" />

    <col width="97" />

    <col width="91" />

    <col width="100" />
  </colgroup>

  <thead>
    <tr>
      <th>Catalog name</th>
      <th>Machine type</th>
      <th>Billed rate/hour</th>
      <th>vCPUs</th>
      <th>Memory</th>
      <th>Sleep</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>`murmur-e2-medium`</td>
      <td>`e2-medium`</td>
      <td>\$0.03685</td>
      <td>2 shared-core</td>
      <td>4 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2-standard-2`</td>
      <td>`n2-standard-2`</td>
      <td>\$0.10681</td>
      <td>2</td>
      <td>8 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2-standard-4`</td>
      <td>`n2-standard-4`</td>
      <td>\$0.21362</td>
      <td>4</td>
      <td>16 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2-standard-8`</td>
      <td>`n2-standard-8`</td>
      <td>\$0.42735</td>
      <td>8</td>
      <td>32 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2-standard-16`</td>
      <td>`n2-standard-16`</td>
      <td>\$0.85459</td>
      <td>16</td>
      <td>64 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2-standard-32`</td>
      <td>`n2-standard-32`</td>
      <td>\$1.70929</td>
      <td>32</td>
      <td>128 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2-standard-64`</td>
      <td>`n2-standard-64`</td>
      <td>\$3.41858</td>
      <td>64</td>
      <td>256 GiB</td>
      <td>Cold</td>
    </tr>

    <tr>
      <td>`murmur-n2d-standard-2`</td>
      <td>`n2d-standard-2`</td>
      <td>\$0.09295</td>
      <td>2</td>
      <td>8 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2d-standard-4`</td>
      <td>`n2d-standard-4`</td>
      <td>\$0.18590</td>
      <td>4</td>
      <td>16 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2d-standard-8`</td>
      <td>`n2d-standard-8`</td>
      <td>\$0.37180</td>
      <td>8</td>
      <td>32 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2d-standard-16`</td>
      <td>`n2d-standard-16`</td>
      <td>\$0.74349</td>
      <td>16</td>
      <td>64 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2d-standard-32`</td>
      <td>`n2d-standard-32`</td>
      <td>\$1.48709</td>
      <td>32</td>
      <td>128 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2d-standard-48`</td>
      <td>`n2d-standard-48`</td>
      <td>\$2.23058</td>
      <td>48</td>
      <td>192 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-n2d-standard-64`</td>
      <td>`n2d-standard-64`</td>
      <td>\$2.97407</td>
      <td>64</td>
      <td>256 GiB</td>
      <td>Cold</td>
    </tr>

    <tr>
      <td>`murmur-t2d-standard-2`</td>
      <td>`t2d-standard-2`</td>
      <td>\$0.09295</td>
      <td>2</td>
      <td>8 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-t2d-standard-4`</td>
      <td>`t2d-standard-4`</td>
      <td>\$0.18590</td>
      <td>4</td>
      <td>16 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-t2d-standard-8`</td>
      <td>`t2d-standard-8`</td>
      <td>\$0.37180</td>
      <td>8</td>
      <td>32 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-t2d-standard-16`</td>
      <td>`t2d-standard-16`</td>
      <td>\$0.74349</td>
      <td>16</td>
      <td>64 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-t2d-standard-32`</td>
      <td>`t2d-standard-32`</td>
      <td>\$1.48709</td>
      <td>32</td>
      <td>128 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-t2d-standard-48`</td>
      <td>`t2d-standard-48`</td>
      <td>\$2.23058</td>
      <td>48</td>
      <td>192 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-t2d-standard-60`</td>
      <td>`t2d-standard-60`</td>
      <td>\$2.78828</td>
      <td>60</td>
      <td>240 GiB</td>
      <td>Cold</td>
    </tr>

    <tr>
      <td>`murmur-c3-standard-44`</td>
      <td>`c3-standard-44`</td>
      <td>\$2.43947</td>
      <td>44</td>
      <td>176 GiB</td>
      <td>Warm</td>
    </tr>
  </tbody>
</table>

### AWS machine types

<table>
  <colgroup>
    <col width="173" />

    <col width="150" />

    <col width="123" />

    <col width="83" />

    <col width="100" />

    <col width="100" />
  </colgroup>

  <thead>
    <tr>
      <th>Catalog name</th>
      <th>Machine type</th>
      <th>Billed rate/hour</th>
      <th>vCPUs</th>
      <th>Memory</th>
      <th>Sleep</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>`murmur-m5-large`</td>
      <td>`m5.large`</td>
      <td>\$0.10560</td>
      <td>2</td>
      <td>8 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-m5-xlarge`</td>
      <td>`m5.xlarge`</td>
      <td>\$0.21120</td>
      <td>4</td>
      <td>16 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-m5-2xlarge`</td>
      <td>`m5.2xlarge`</td>
      <td>\$0.42240</td>
      <td>8</td>
      <td>32 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-m5-4xlarge`</td>
      <td>`m5.4xlarge`</td>
      <td>\$0.84480</td>
      <td>16</td>
      <td>64 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-m5-8xlarge`</td>
      <td>`m5.8xlarge`</td>
      <td>\$1.68960</td>
      <td>32</td>
      <td>128 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-m5-16xlarge`</td>
      <td>`m5.16xlarge`</td>
      <td>\$3.37920</td>
      <td>64</td>
      <td>256 GiB</td>
      <td>Cold</td>
    </tr>

    <tr>
      <td>`murmur-c5-large`</td>
      <td>`c5.large`</td>
      <td>\$0.09350</td>
      <td>2</td>
      <td>4 GiB</td>
      <td>Warm</td>
    </tr>

    <tr>
      <td>`murmur-c5-9xlarge`</td>
      <td>`c5.9xlarge`</td>
      <td>\$1.68300</td>
      <td>36</td>
      <td>72 GiB</td>
      <td>Warm</td>
    </tr>
  </tbody>
</table>

## NAT IP

The VM's public IPv4 address is billed at \$0.0055 per hour, accrued per VM-second while the VM is running, and stops when the VM stops or sleeps.

## Storage

Storage costs cover three separate parts, combined into one Storage line item. [Boot disk](#boot-disk) and [sleep memory image](#sleep-memory-image) are priced from the [disk type](#disk-type-rates); [baked images](#baked-images) have their own rate.

<table>
  <colgroup>
    <col width="183" />

    <col width="276" />

    <col width="275" />
  </colgroup>

  <thead>
    <tr>
      <th>Component</th>
      <th>Charged while</th>
      <th>Rate</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>[Boot disk](#boot-disk)</td>
      <td>The VM's disk exists, whether the VM runs or sleeps</td>
      <td>Disk type rate × provisioned GiB</td>
    </tr>

    <tr>
      <td>[Sleep memory image](#sleep-memory-image)</td>
      <td>A machine type that preserves memory on sleep holds one</td>
      <td>Disk type rate × machine RAM</td>
    </tr>

    <tr>
      <td>[Baked image](#baked-images)</td>
      <td>You keep a custom VM image</td>
      <td>Per stored GiB-hour</td>
    </tr>
  </tbody>
</table>

### Boot disk

The boot disk is the VM's root filesystem. It accrues per VM-second from VM creation until the disk is deleted, whether the VM is running or sleeping.

### Sleep memory image

The sleep memory image holds the machine's RAM so a warm VM can resume with memory intact. It is sized to the machine's RAM, and behavior differs by cloud:

* **AWS** — a memory-image volume of the selected disk type exists for the VM's full lifetime, whether the VM is running or sleeping.
* **GCP** — a memory snapshot exists only while the VM is suspended.

Machine types marked **Cold** preserve no memory and incur no sleep-memory charge.

### Baked images

Baked images are billed at \$0.000077 per stored GiB-hour, based on the image's stored size (the compressed archive or EBS snapshot, not the nominal size of the source disk), so one GiB kept for a 730-hour month costs \$0.05621. Deleting an unused image stops the charge.

### Disk type rates

Disk type rates price the [boot disk](#boot-disk) and [sleep memory image](#sleep-memory-image). The [disk-type reference](/catalog/disk-type) lists the built-in entries.

These rates are quoted per GiB-month and metered by the second, prorated by 730 hours per month: one GiB of `murmur-pd-balanced` held for six hours costs \$0.1100 / 730 × 6, or \$0.0009.

<table>
  <colgroup>
    <col width="147" />

    <col width="136" />

    <col width="198" />

    <col width="139" />
  </colgroup>

  <thead>
    <tr>
      <th>Catalog name</th>
      <th>Disk type</th>
      <th>Priced regions</th>
      <th>Billed rate</th>
    </tr>
  </thead>

  <tbody>
    <tr>
      <td>`murmur-pd-standard`</td>
      <td>GCP `pd-standard`</td>
      <td>us-central1, us-east1, us-west1</td>
      <td>\$0.0440 per GiB-month</td>
    </tr>

    <tr>
      <td>`murmur-pd-balanced`</td>
      <td>GCP `pd-balanced`</td>
      <td>us-central1, us-east1, us-west1</td>
      <td>\$0.1100 per GiB-month</td>
    </tr>

    <tr>
      <td>`murmur-pd-ssd`</td>
      <td>GCP `pd-ssd`</td>
      <td>us-central1, us-east1, us-west1</td>
      <td>\$0.1870 per GiB-month</td>
    </tr>

    <tr>
      <td>`murmur-gp3`</td>
      <td>AWS `gp3`</td>
      <td>us-east-1, us-west-2</td>
      <td>\$0.0880 per GiB-month</td>
    </tr>
  </tbody>
</table>

### Machine and disk types you define

When an admin sets up a [customer-managed placement](/guides/customer-placements) in a region the built-ins do not price, they add their own [machine-type](/catalog/machine-type) and [disk-type](/catalog/disk-type) entries. VMs allocated from those entries are priced from the rates those entries carry for the VM's region:

* `cost_per_hour` on the machine type
* `cost_per_gib_month` on the disk type
* `performance_cost_per_month` on the disk type — a flat per-disk charge that does not scale with size, and applies to the boot disk only

Price every region your placements use in both entries. Enter the rate your cloud provider charges you. Because the placement is customer-managed, Murmur bills 10% of the total these rates produce, and your cloud provider bills you for the VM itself.

## LLM Inference

<Info>
  **Bring your own model credential to skip LLM Inference charges**

  Run [`murmur setup`](/cli/setup) to register your own provider API key or subscription token (Anthropic, OpenAI, Claude, Codex, etc.) as a [developer secret](/concepts/secrets#developer-secrets). Agents you spawn use it instead of Murmur's credential.

  Agents that do not run as you — a [flight](/concepts/flights), or anything else started by automation — run under a [service profile](/catalog/service-profile) and take their credential from that profile's `anthropic_api_key_secret`, `claude_oauth_token_secret`, or `openai_api_key_secret`. Set those too, or those calls stay on Murmur's credential.
</Info>

An agent runs on a **Murmur-provided model credential** if it has no credentials of its own. These LLM Inference calls are billed on four token counts, valued at the [published per-token rates](https://docs.macroscope.com/model-pricing#pricing) for the model that served the call:

* uncached input tokens
* cache-read tokens
* cache-write tokens
* output tokens

Those per-token rates are the one place this page does not quote a final price. The charge is that token subtotal plus a 10% markup. Per-model rates are not listed on this page — [the rate page](https://docs.macroscope.com/model-pricing#pricing) carries a figure for each of the four counts above, model by model.

<Note>
  Calls on your own credentials are still recorded — token counts reach the dashboard's usage views and the per-turn cost figure exactly as Murmur-provided calls do.

  The per-turn cost the dashboard shows (and the cumulative API cost in [`murmur ls`](/cli/ls) and [`murmur status`](/cli/status)) is a **list-price estimate**, not a charge. It values the turn's tokens at the provider's published rates, with cache-TTL, service-tier, fast-mode, and pinned-region multipliers applied per call. Use it to compare turns, agents, and models, not to predict your bill.
</Note>

## Historical Session Storage

Each agent execution with a persisted session keeps one JSONL transcript in Murmur-managed session storage:

* The coding backend's conversation state, used to restore or [resurrect](/cli/spawn#resurrect-a-previous-session) the execution.
* Downloadable from the agent view.
* Separate from the dashboard's timeline and event history.
* Survives the execution finishing, sleeping, being killed, or losing its VM.

Stored bytes are metered once a day at \$0.00022 per decimal MB-month — approximately \$0.2362 per GiB-month — prorated for the fraction of your billing period the transcript existed.

To stop future accrual, permanently purge the stopped agent with [`murmur delete`](/cli/delete) — this covers every execution and descendant agent, and every version of their transcripts. Preview the scope with `murmur delete --dry-run <slug>`, and [`murmur kill`](/cli/kill) first if anything in the subtree is still running.

<Note>
  There is no separate retention setting or workspace-wide session purge. Deleting a workspace catalog entry does not delete transcripts; deleting the tenant purges all session objects under it.
</Note>