cancel-job
ChatGPTCancel a Serverless job that is queued or in progress.
Overview
Runpod is a GPU cloud for AI workloads. This plugin manages your Runpod account directly: browse GPUs and live stock, launch and control Pods, deploy and run Serverless endpoints, and keep track of what it all costs. What you can do: - Find the right GPU: list GPU and CPU types with prices, and check live stock by data center and CUDA version. - Run Pods: create, start, stop, restart, update and delete GPU or CPU Pods, and read their logs. - Deploy Serverless endpoints: create endpoints from your own image or a Runpod Hub repo, submit jobs, poll results, stream output, and check worker health and logs. - Manage storage and setup: network volumes, global volumes, templates and SSH keys. - Run Instant Clusters of multi-GPU Pods. - Review spend: billing history for Pods, Serverless, storage, clusters and Public Endpoints. - Get answers: search the Runpod docs and look up how Runpod works. You sign in with your Runpod account through OAuth, so no API key is stored.
Tools
create-cluster
ChatGPTCreate an Instant Cluster: compute.podCount identical Pods joined by a high-speed interconnect, each with compute.gpuCountPerPod GPUs of compute.gpuTypeId. Billable from creation for every Pod in the cluster, at the GPU type's hourly price times the total GPU count. The compute shape and type are fixed at creation; only the name can change later. The container configuration applies to every Pod, and templateId is the only way to use a private image; a bare registry property is rejected. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overview
create-delegation
ChatGPTRegister an ECR delegation. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overview
create-endpoint
ChatGPTCreate a Serverless endpoint. Billable while its workers run, at the worker's hourly price. Worker hardware is either the gpu object (gpu.pools takes Serverless pool ids such as ADA_24, and gpu.count sets GPUs per worker) or the cpu array of CPU flavor configurations. type and scaling are required and set the request-routing model. Workers cold-start on their first job, which commonly takes minutes. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overview
create-global-volume
ChatGPTCreate a global volume. Provisions a new global volume — storage that mounts into workloads in any data center. Only a
name is required; storage is provisioned for you. Accounts have a cap on how many global volumes they can hold; creating one past it fails with 400. Returns 201 with the created global volume, including its assigned id. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overviewcreate-network-volume
ChatGPTCreate a network volume. Provisions a new network volume — persistent, network-attached storage that can be mounted into pods and serverless workers. Required inputs are
name, size (in GB), and dataCenter; an optional type selects the storage tier and is immutable after creation. See CreateNetworkVolumeRequest for the size bounds and tier options. This creates a billable persistent resource that keeps incurring storage charges until it is deleted. Returns 201 with the created network volume, including its assigned id. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overviewcreate-pod
ChatGPTCreate a GPU or CPU Pod. Billable from creation until it is stopped or terminated. GPU selection is the gpu object: gpu.id takes a GPU type id such as "NVIDIA GeForce RTX 4090", and gpu.count sets the number of GPUs. Cloud tier and data center choices limit which machines can host it. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overview
create-template
ChatGPTCreate a template. Creates a reusable container-configuration preset — image, disk, ports, env, registry, and mount settings — for pods and serverless endpoints. Pass its ID as
templateId to createPod or createEndpoint, or spread its fields into the request body directly. Returns the created template. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overviewdelete-cluster
ChatGPTDelete an Instant Cluster permanently. Every member Pod is terminated and its container disk is lost.
delete-endpoint
ChatGPTDelete a Serverless endpoint permanently and immediately.
delete-global-volume
ChatGPTDelete a global volume. Permanently deletes a global volume and its stored data. Fails with
409 while the volume is still mounted: terminate every pod that mounts it (stopped pods count) and detach it from every serverless endpoint first. Nothing is detached or terminated on your behalf.delete-network-volume
ChatGPTDelete a network volume. Permanently deletes a network volume and releases its storage.
delete-pod
ChatGPTTerminate a Pod permanently. Its container disk is lost; data on an attached network volume is kept. Stopping a Pod instead releases its compute but keeps its disk.
delete-template
ChatGPTDelete a template permanently. An endpoint that references the template can stop working once it is deleted.
deploy-hub-repo
ChatGPTDeploy a Runpod Hub repo's listed release as a new Serverless endpoint, the same as clicking Deploy on the Hub. Identifies the repo by repo ("owner/name") or hubReleaseId. The release supplies the image, container disk, CUDA constraints and environment-variable defaults; env overrides or fills in values, and keys the release requires without a default have to be supplied. GPUs come from the release config when it sets them, otherwise from gpuIds (pool names such as 'ADA_24'). Billable while the endpoint's workers run.
endpoint-health
ChatGPTGet the health rollup for a Serverless endpoint: worker counts by state and job queue statistics. The rollup can lag the per-worker state, and a worker whose container keeps failing to start can still report RUNNING or THROTTLED here. Zero workers means the endpoint is waiting for GPU capacity.
get-capacity
ChatGPTGPU capacity across host CUDA versions, as a matrix. By default one call returns, for each GPU type, overall stock plus AVAILABLE or UNAVAILABLE per host CUDA version. With cudaVersions it checks each listed version and returns graded stock (High, Medium, Low, Out) and the lowest on-demand price per version; a check that fails transiently reports an error for that version only. Out-of-stock GPUs are included unless includeUnavailable is false. Live public catalog data, so page cursors can shift between calls.
get-cluster
ChatGPTGet a cluster. Returns a single cluster by ID. The pods field is an aggregate summary (total + count by status); fetch the member pods themselves from /v2/clusters/{id}/pods.
get-cpu-type
ChatGPTGet a CPU type. Returns a single CPU type with pricing. Availability details are included only when requested with include=AVAILABILITY, which requires
product — stock differs by product context.get-data-center
ChatGPTGet a data center. Returns a single data center. Availability is included only when requested with include=GPU_AVAILABILITY or include=CPU_AVAILABILITY.
get-endpoint
ChatGPTGet a serverless endpoint. Returns a single serverless endpoint by ID.
get-endpoint-build
ChatGPTGet a serverless endpoint build. Returns one of the endpoint's GitHub builds by id, regardless of age — no need to page through
GET /v2/serverless/{id}/builds to reach it.get-global-volume
ChatGPTGet a global volume. Returns a single global volume by ID.
get-gpu-type
ChatGPTGet a GPU type. Returns a single GPU type with pricing. Availability details are included only when requested with include=AVAILABILITY, which requires
product — stock differs by product context.get-job-status
ChatGPTGet the status of a Serverless job, and its output once it completes. Statuses: IN_QUEUE, IN_PROGRESS, COMPLETED, FAILED, CANCELLED, TIMED_OUT. With wait (milliseconds, up to 45000) the server polls until the job reaches a final status or the wait runs out, in which case it returns the latest status with pollingTimedOut: true; without wait it returns the current status at once. An IN_QUEUE result includes a workerHealth summary of the endpoint's workers.
get-network-volume
ChatGPTGet a network volume. Returns a single network volume by ID.
get-organization
ChatGPTGet your current organization. Returns the authenticated caller's current organization. Organization context comes from the caller's membership; no organization selector is required. Returns 404 when the caller does not belong to an organization, or 403 when organizations are unavailable to the caller.
get-organization-user
ChatGPTGet an organization member. Returns one member of the authenticated caller's organization by user ID, with the role they hold. Any member can read it. A user who is not a member, including one from another organization, returns 404. Returns 404 when the caller does not belong to an organization, or 403 when organizations are unavailable to the caller.
get-pod
ChatGPTGet a pod. Returns a single pod by ID.
get-ssh-keys
ChatGPTList registered SSH public keys. Returns the account's registered SSH public keys — the keys provisioned into pods created with
startSsh and used to authenticate the SSH connections reported in a pod's ssh block.get-template
ChatGPTGet a template. Returns the full configuration of a single template by ID. Serves both templates you own and public catalog templates — everything you can read. Updates and deletes remain restricted to templates you own.
list-audit-logs
ChatGPTList the account's audit trail, newest first. Results are cursor-paginated and record the actor, action, resource, timestamp, and change details. Filters include actor IDs, actions, time range, and text search. The account's retention window and plan determine which entries are available.
list-billing
ChatGPTGet aggregated billing history. Returns time-bucketed total spend across all billable Runpod resources for the authenticated user. Use startTime/endTime with bucketSize for an explicit range, or lastN with bucketSize for the most recent buckets. Each record reports one bucket's total plus pod, serverless, storage, public endpoint, and Instant Cluster cost components. The metadata block echoes the resolved query window, record count, and totals across all returned buckets.
list-cluster-billing
ChatGPTGet cluster billing history. Returns Cluster billing history for the authenticated user, split into time buckets by startTime/endTime with bucketSize or by lastN recent buckets. Use clusterId to filter to one cluster; without it, records are emitted per cluster per bucket. Each record includes GPU compute, disk, inter-node networking, and total amounts. Clusters are GPU-only, so no CPU cost component is returned.
list-cluster-pods
ChatGPTList a cluster's pods. Returns the full member pods of a cluster. The cluster summary (
GET /v2/clusters/{id}) carries only aggregate pod counts; this endpoint returns each member as a complete Pod object.list-clusters
ChatGPTList clusters. Returns clusters owned by the authenticated user, cursor-paginated newest-first; an omitted
limit defaults to 1000. Follow pagination.nextCursor until hasNextPage is false.list-cpu-types
ChatGPTList CPU types. Returns available CPU flavors. Availability is included only when requested with include=AVAILABILITY, which requires
product — stock differs by product context.list-data-centers
ChatGPTList data centers. Returns available data center locations with region, compliance, supported network volume tiers, and global networking support. Use include=GPU_AVAILABILITY or include=CPU_AVAILABILITY to add per-resource availability arrays to each data center. The regions, networkVolumeTypes, compliance, and globalNetwork query parameters filter the list before it is returned.
list-delegations
ChatGPTList all ECR delegations
list-endpoint-billing
ChatGPTGet public endpoint billing history. Returns Runpod public endpoint billing history for the authenticated user, split into time buckets by startTime/endTime with bucketSize or by lastN recent buckets. Each record reports the endpoint total for one bucket, and metadata echoes the resolved query window, record count, and total endpoint amount across all returned records.
list-endpoint-builds
ChatGPTList serverless endpoint builds. Returns the endpoint's GitHub build history, newest first (Runpod GitHub-build integration), cursor-paginated; an omitted
limit defaults to 100, so a bare request returns at most the 100 most recent builds. Any build can also be fetched by id via GET /v2/serverless/{id}/builds/{buildId}. Stream a build's logs via /v2/serverless/{id}/builds/{buildId}/logs.list-endpoint-releases
ChatGPTList serverless endpoint releases. Returns the endpoint's release history (newest first) plus a rollout summary of how many workers are running the current version. Each release is a versioned configuration snapshot with a
diff of what changed; build-driven releases carry a buildId (fetch build detail via the builds sub-routes). Releases are cursor-paginated newest-first; an omitted limit defaults to 1000. The rollout summary always describes the endpoint's current state, independent of the page requested.list-endpoint-workers
ChatGPTList serverless endpoint workers. Lists the active workers for a serverless endpoint. Returns. A
200 with a ListEndpointWorkersResponse: a workers array (one entry per active worker, each carrying its id, status, and runtime details) plus a summary of worker counts grouped by status. Only currently active workers are included; scaled-down workers are not returned. How `status` is determined. Each worker's status is derived by reconciling the worker pod's lifecycle status with the endpoint's live job-queue view (which workers are actively serving requests). When the job-queue view is unavailable, the response degrades gracefully: the shape is unchanged, but each status and the summary counts fall back to pod lifecycle alone.list-endpoints
ChatGPTList the Serverless endpoints the account owns, trimmed to id, name, image, compute config, worker counts, scaling and data centers; environment variables and request URLs are left out. Returns one page at a time; cursor=nextCursor returns the next page.
list-global-volumes
ChatGPTList global volumes. Returns global volumes owned by the authenticated user, cursor-paginated newest-first; an omitted
limit defaults to 1000. Follow pagination.nextCursor until hasNextPage is false.list-gpu-types
ChatGPTList GPU types. Returns available GPU types with pricing. Availability is included only when requested with include=AVAILABILITY, which requires
product — stock differs by product context. With countryCodes, the list is narrowed to GPU types deployable in those countries, so "this geography + this chip" resolves in one read.list-hub-repos
ChatGPTList repos published to the Runpod Hub (prebuilt Serverless workers and Pod templates, such as vLLM and ComfyUI), sorted by deploy count. Public catalog; no sign-in needed. Each result includes the listed release with its hubReleaseId and image name. includeConfig: true adds each release's config (hardware requirements and environment-variable schema), which is large; repoOwner and searchTerm narrow the list.
list-network-volume-billing
ChatGPTGet network volume billing history. Returns network volume billing history for the authenticated user, split into time buckets by startTime/endTime with bucketSize or by lastN recent buckets. Use networkVolumeId to filter to one volume; without it, records are emitted per volume per bucket. Each record includes standard storage, high-performance storage, and total amounts, while metadata reports the resolved query, distinct volume count, and totals across the returned records.
list-network-volumes
ChatGPTList network volumes. Returns network volumes owned by the authenticated user, cursor-paginated newest-first; an omitted
limit defaults to 1000. Follow pagination.nextCursor until hasNextPage is false.list-organization-users
ChatGPTList organization members. Returns every member of the authenticated caller's organization with the role each holds. Any member can read the roster. The list is complete and takes no pagination parameters. Returns 404 when the caller does not belong to an organization, or 403 when organizations are unavailable to the caller.
list-pod-billing
ChatGPTGet pod billing history. Returns pod-only billing detail for the authenticated user, split into time buckets by startTime/endTime with bucketSize or by lastN recent buckets. Use podId to narrow the response to one GPU or CPU pod; without it, records are emitted per pod per bucket. Each record includes podId, GPU, CPU, disk, and total amounts, while metadata echoes the resolved query and totals across the pod records. Use listBilling when you need aggregate spend across every billable resource family.
list-pods
ChatGPTList pods. Returns pods owned by the authenticated user. Cluster member pods are excluded by default; set
includeClusterPods=true to include them (each carries a non-null cluster membership block). Results are cursor-paginated newest-first; an omitted limit defaults to 1000. The cluster-member exclusion is applied before the page is cut, so a page holds limit pods whenever that many remain. Follow pagination.nextCursor until hasNextPage is false.list-public-endpoints
ChatGPTList Runpod Public Endpoints: managed, pay-per-use model APIs (text, image, video, audio) that need no deployment. Public catalog; no sign-in needed. Each result includes its endpointId, model name, modality, owner and pricing. Only live endpoints are listed unless includeOffline is true.
list-public-templates
ChatGPTList public templates. Returns the public template catalog.
source selects which slice: official (the default) is Runpod-curated templates, verified is community templates Runpod has verified, and community is everything else other users have shared publicly. Both pod and serverless templates appear — use each entry's serverless flag to tell them apart. registry is always null for templates you don't own. Your own templates (public or private) are managed under /v2/templates; fetch any individual template — catalog or owned — via /v2/templates/{id}. At most 100 templates are returned. Cursor pagination is not yet supported here; pagination is always the exhausted marker (nextCursor: null, hasNextPage: false).list-serverless-billing
ChatGPTGet Serverless billing history, split into time buckets. By default the response has one record per endpoint per bucket, which can be very large on accounts with many endpoints; lastN: 1 (or one bucket spanning the whole window) returns one record per endpoint, and serverlessId narrows it to a single endpoint. Pod billing is reported separately.
list-templates
ChatGPTList the templates visible to the account, with id, name, image and Serverless flag for each. Returns 20 templates per page by default; pagination.nextCursor returns the next page, and a template missing from one page can be on a later one.
lookup-concept
ChatGPTLook up one Runpod concept (Pod, network volume, machine, worker, template and so on) by id, name or alias; case, plurals and small typos are tolerated, and matched_as shows the key used when the match was not exact. Returns its summary, fields, states, relations and rules, each rule with its public source, plus the concepts that point at it. Answers exact questions such as whether a volume disk is tied to one machine.
purge-endpoint-queue
ChatGPTRemove all queued jobs from a Serverless endpoint. Jobs already in progress keep running.
restart-pod
ChatGPTRestart a running Pod's container in place. Whatever is running in the container is interrupted and the container disk is wiped; the volume disk and any attached network volume keep their data. Returns the updated Pod; a Pod that is not RUNNING returns 409.
retry-job
ChatGPTRetry a failed or timed-out Serverless job. Only works for jobs with FAILED or TIMED_OUT status. The previous output is removed and the job is requeued.
revoke-delegation
ChatGPTRevoke an ECR delegation
run-endpoint
ChatGPTSubmit an asynchronous job to a Serverless endpoint. Returns a job ID immediately; the result stays available for 30 minutes after the job completes.
runsync-endpoint
ChatGPTSubmit a synchronous job to a Serverless endpoint and wait up to 45000 ms for the result. If the job outlives the wait, the response carries the job ID and its current status (IN_QUEUE or IN_PROGRESS). Maximum payload 20 MB; results expire 1 minute after completion.
search-concepts
ChatGPTSearch the rules of the Runpod concept graph by words, for example "restart stopped pod no gpu" or "network volume data center". Returns the best-matching rules with their concept and public evidence.
search-docs
ChatGPTSearch Runpod documentation, curated support answers, the Runpod blog and runpod.io for passages that answer a question, for example "how do I attach a network volume to a Serverless endpoint". Returns ranked passages with their source URLs. Covers how-to and product questions, not account or resource state.
set-endpoint-gpus
ChatGPTSet which GPUs a Serverless endpoint's workers run on, including pinning specific GPU types. Takes either a raw gpuIds string, or pools plus optional excludeGpuTypeIds, from which the exclusion string is built: a pool allows every GPU type in it, and excluding all but one pins that type. Changes only GPU settings and keeps the endpoint's other settings.
start-pod
ChatGPTStart a stopped Pod (status EXITED or ERROR). Billable: the Pod accrues its hourly price while it runs. Returns the updated Pod; a Pod in any other status returns 409.
stop-pod
ChatGPTStop a running or provisioning Pod. It releases its GPU or CPU and stops compute billing; the container disk is wiped, while the volume disk and any attached network volume keep their data and volume storage keeps billing. Returns the updated Pod; a Pod in any other status returns 409.
stream-job
ChatGPTCollect streaming output from a Serverless job whose worker streams its output. Polls the job's stream until it reaches COMPLETED, FAILED, CANCELLED or TIMED_OUT, for up to 45 seconds; if time runs out first, it returns the chunks collected so far with pollingTimedOut: true, and a later call resumes where it left off.
stream-pod-logs
ChatGPTRead a bounded snapshot of a Pod's logs (container and/or system source). Holds the live stream open for maxWaitMs and returns the parsed log lines;
truncated: true means the byte cap cut the output. Use since to resume from a timestamp.stream-worker-logs
ChatGPTRead a bounded snapshot of a Serverless worker's logs (container, system, or both). Holds the live stream open for maxWaitMs and returns the parsed log lines; truncated: true means the byte cap cut the output. A worker whose container keeps failing to start can report RUNNING or THROTTLED while its logs show the failure.
update-cluster
ChatGPTRename a cluster. Renames a cluster. This endpoint only changes the cluster name — compute shape, type, and container configuration are fixed at creation and cannot be updated. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overview
update-endpoint
ChatGPTPartially update a Serverless endpoint (PATCH): only fields present in the body change. gpu.pools and gpu.excludedTypes are one selection, so sending pools without excludedTypes clears existing exclusions; an exclusion naming a type outside pools returns 400, and a CUDA-only or count-only change does not need pools. Mutable fields: name, gpu, cpu, workers (min/max/idleTimeout), scaling (type plus queueDelay or requestCount), dataCenterIds, networkVolumes, timeout, flashboot, and the container settings (image, args, disk, ports, env, registry). cpu replaces a CPU endpoint's whole flavor selection. The compute family is fixed: gpu on a CPU endpoint, cpu on a GPU one, or both returns 400. Worker, scaling and timeout changes apply promptly; container changes (image, env) create a new release that rolls out as workers cycle, so in-flight workers keep the previous version. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overview
update-global-volume
ChatGPTUpdate a global volume. Updates mutable fields on a global volume. Only provided fields are changed. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overview
update-network-volume
ChatGPTUpdate a network volume. Updates mutable fields on a network volume. Only provided fields are changed. Note:
size may only increase; attempts to reduce size will be rejected. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overviewupdate-pod
ChatGPTUpdate a pod. Partially updates a pod's configuration. This is a PATCH: only the fields present in the body are changed, and omitted fields are left untouched. Use empty values only when you explicitly mean to clear a field (for example, set
registry to null or set ports to []). See UpdatePodRequest for the full body. Mutable fields: name, image, args, disk, ports, env, registry, mounts, locked, and globalNetworking. Some changes apply immediately while others (e.g. globalNetworking) take effect on the pod's next start/restart, as noted on the individual fields. Pods that belong to a Cluster cannot be updated here — manage them through /v2/clusters/{id}. Returns 200 with the full updated pod. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overviewupdate-ssh-keys
ChatGPTReplace registered SSH public keys. Replaces the account's full set of registered SSH public keys. Existing keys not present in the request are removed; send
[] to remove all keys. Keys take effect for pods created afterwards with startSsh — running pods are not updated. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overviewupdate-template
ChatGPTUpdate a template. Partially updates a template. This is a PATCH: only the fields present in the body are changed; omitted fields are left untouched. See
UpdateTemplateRequest for the full body. Mutable fields: name, image, args, disk, ports, env, registry, mounts, serverless, public, and category. Only the template's owner can update it (authenticated via the request's API key); public catalog templates are readable via GET but return 404 here. Returns 200 with the full updated template. Pods and endpoints already created from this template are not changed retroactively — the template is a snapshot applied at creation time. Calls the Runpod REST API v2: https://docs.runpod.io/api-reference-v2/overviewCapabilities
Writes
Example Prompts
Click any prompt to copy it.
