Version Sets & Promotion
Version sets define which agent, state machine, and model preferences a service uses. They enable safe iteration like Git branches while keeping production stable.
Purpose: Pin exact versions for a service run (agent, state machine, LLM model preferences).
Analogy: Treat version sets like Git branches for your service configuration.
Two special sets exist on every service:
edge: Always points to the latest agent and state machine with no model preferences. Locked from updates; cannot be deleted.release: The default version set used when clients don’t specify one. Cannot be deleted; can be updated to promote changes.
Guardrails
edgecannot be updated or deleted.releasecannot be deleted and is the default when no version set is specified.Custom version sets cannot be deleted if referenced by simulation unit tests.
What Each Special Set Means
edge(locked default settings)Always latest agent/state machine; no explicit LLM preferences.
Cannot be updated via the upsert API and cannot be deleted.
Use for quick local validation or smoke checks; not for stable user exposure.
release(default exposed to users)Used by conversations and real-time APIs when
service_version_set_nameis not provided.Cannot be deleted, but can be updated to promote a tested configuration to production.
Creating and Updating Version Sets
Create/Update: Use the Service Version Set upsert API to create named sets (e.g.,
personal-alex,test,preview). All sets exceptedgeare updatable.Delete: Allowed for custom sets, but not for
edgeorrelease. Deletion is blocked if a set is referenced by any simulation unit tests.Permissions:
Create:
Service:CreateVersionSetUpdate:
Service:UpdateVersionSetDelete:
Service:DeleteVersionSet
How Version Sets Are Used at Runtime
Conversation APIs default to
releaseifservice_version_set_nameis omitted.Simulation unit tests and test runs must explicitly specify the target version set.
Services are created with both
edgeandreleaseby default;releasewill equaledgeunless specified otherwise at creation time.
Recommended Workflow (Git-Branch Style)
Personal branch
Create a personal version set (e.g.,
personal-yourname), pin agent/state machine versions and set LLM preferences.Iterate locally; validate via small focused simulation unit tests.
Merge into
test
Upsert the shared
testversion set with your candidate configuration.Run your simulation unit test set on
test. Fix regressions here.
Promote to
preview
Copy the
testconfiguration intopreview(another named version set).Run the full simulation suite on
previewand conduct stakeholder UAT.
Release
If
previewpasses full sims and UAT, promote by updatingreleasewith the same configuration (i.e., upsertreleaseto matchpreview).Do not delete
releaseoredge. Use them as stable anchors.
Promotion Flow
Practical Tips
Naming Conventions
Use lowercase letters, digits,
-or_only.Don’t use
/,., or names starting with$.Examples:
personal-alex,test,preview,release.Drift control: Treat
releaseas immutable except during deliberate promotion. Keep a changelog of which version set was promoted, when, and by whom.Safety checks: Block deletion of sets referenced by simulations. Prefer deprecation and migration over deletion when many artifacts depend on a set.
Fast rollback: Keep the last known good configuration for
releasehandy (export JSON or maintain a dedicatedrelease-previousset) to enable quick rollback.Observability: Compare simulation metrics between
test,preview, andreleaseto quantify impact before promotion.
API Touchpoints (Operational Behavior)
API References
Create a new service. Depending on whether an active service with the same name already exists, the endpoint behaves differently:
- If
is_activeisFalse, creates an inactive new service. - If
is_activeisTrueand no active service with the given name exists, creates a new active service. - If
is_activeisTrueand an active service with the given name exists, this endpoint throws an error.
The new service will automatically contain an edge version set that uses the latest Agent and ServiceHierarchicalStateMachine versions with no LLM model preference.
It will also create a release version set, that will equal to what's specified in the request if the release_version_set is specified, or equal to edge if not.
Permissions
This endpoint requires the following permissions:
Service:CreateServicefor the service to create.Service:CreateVersionSetfor theedgeandreleaseversion sets.
Amigo issued JWT token that identifies an user. It's issued either after logging in through the frontend, or manually through the SignInWithAPIKey endpoint.
An optional organization identifier that indicates from which organization the token is issued. This is used in rare cases where the user to authenticate is making a request for resources in another organization.
The Mongo cluster name to perform this request in. This is usually not needed unless the organization does not exist yet in the Amigo organization infra config database.
[]The ID of the state machine that this service uses.
^[a-f0-9]{24}$The ID of the agent that this service uses.
^[a-f0-9]{24}$The name of this service.
A description of this service.
Whether the newly-created service is active. Only active services are visible to users on the dashboard. You can later adjust the activeness of this service.
The release version set to use for this service. If not specified, the release version set will be the same as the edge version set, which uses the
latest agent and state machine versions with no model preference.
Succeeded.
This error could be thrown due to the following reasons:
- The specified ID for agent or service hierarchical state machine versions in the
releaseversion set does not belong to the agent or service hierarchical state machine of the service. - The specified agent or state machine doesn't have any versions.
- The specified LLM config is invalid.
Invalid authorization credentials.
Missing required permissions.
The specified organization, agent, or service hierarchical state machine do not exist.
An active service with the given name already exists.
Invalid request path parameter or request body failed validation.
The user has exceeded the rate limit of 20 requests per minute for this endpoint.
The service is going through temporary maintenance.
POST /v1/{organization}/service/ HTTP/1.1
Host: api.amigo.ai
Authorization: Bearer YOUR_SECRET_TOKEN
X-ORG-ID: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 403
{
"service_hierarchical_state_machine_id": "text",
"agent_id": "text",
"name": "text",
"description": "text",
"is_active": true,
"release_version_set": {
"agent_version_number": 1,
"service_hierarchical_state_machine_version_number": 1,
"llm_model_preferences": {
"ANY_ADDITIONAL_PROPERTY": {
"llm_name": "azure_gpt-4.1-2025-04-14",
"params": {
"ANY_ADDITIONAL_PROPERTY": "anything"
}
}
}
},
"tags": {
"ANY_ADDITIONAL_PROPERTY": "text"
}
}{
"id": "text"
}Upsert a service version set
Upsert a version set for the specified service. Replace the existing version set with the same name if any, or create a new one.
Note that the edge version set cannot be updated.
Permissions
This endpoint may require the following permissions:
Service:CreateVersionSetif the version set does not exist.Service:UpdateVersionSetif the version set already exists.
Amigo issued JWT token that identifies an user. It's issued either after logging in through the frontend, or manually through the SignInWithAPIKey endpoint.
An optional organization identifier that indicates from which organization the token is issued. This is used in rare cases where the user to authenticate is making a request for resources in another organization.
Identifier of the service.
^[a-f0-9]{24}$Name of the version set to upsert.
The Mongo cluster name to perform this request in. This is usually not needed unless the organization does not exist yet in the Amigo organization infra config database.
[]Succeeded.
The agent or service hierarchical state machine version in the request do not belong to the agent or state machine of the service, or attempted to update the edge version set, or the supplied LLM config is invalid.
Invalid authorization credentials.
Missing required permissions.
The specified organization or service do not exist.
Invalid request path parameter or request body failed validation.
The user has exceeded the rate limit of 30 requests per minute for this endpoint.
The service is going through temporary maintenance.
PUT /v1/{organization}/service/{service_id}/version_sets/{version_set_name}/ HTTP/1.1
Host: api.amigo.ai
Authorization: Bearer YOUR_SECRET_TOKEN
X-ORG-ID: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 236
{
"version_set": {
"agent_version_number": 1,
"service_hierarchical_state_machine_version_number": 1,
"llm_model_preferences": {
"ANY_ADDITIONAL_PROPERTY": {
"llm_name": "azure_gpt-4.1-2025-04-14",
"params": {
"ANY_ADDITIONAL_PROPERTY": "anything"
}
}
}
}
}No content
Delete a service version set
Delete the given verion set from the given service.
This endpoint will error if the version set is used in any simulation unit tests.
Permissions
This endpoint requires the following permissions:
Service:DeleteVersionSetfor the version set.
Amigo issued JWT token that identifies an user. It's issued either after logging in through the frontend, or manually through the SignInWithAPIKey endpoint.
An optional organization identifier that indicates from which organization the token is issued. This is used in rare cases where the user to authenticate is making a request for resources in another organization.
Identifier of the service.
^[a-f0-9]{24}$Name of the version set.
The Mongo cluster name to perform this request in. This is usually not needed unless the organization does not exist yet in the Amigo organization infra config database.
[]Succeeded.
The specified version set is used in a simulation unit test, or is the edge or release version set.
Invalid authorization credentials.
Missing required permissions.
The specified organization, service, or version set do not exist.
Invalid request path parameter failed validation.
The user has exceeded the rate limit of 30 requests per minute for this endpoint.
The service is going through temporary maintenance.
DELETE /v1/{organization}/service/{service_id}/version_sets/{version_set_name}/ HTTP/1.1
Host: api.amigo.ai
Authorization: Bearer YOUR_SECRET_TOKEN
X-ORG-ID: YOUR_API_KEY
Accept: */*
No content
- Service creation initializes `edge` and `release`. - Upsert version set API cannot update `edge` but can update `release` and custom sets. - Delete version set API forbids deleting `edge` and `release`, and blocks deletion if a set is used in simulation unit tests. - Conversation and real-time endpoints default to `release` when a version set isn’t specified by the client.
This model lets teams branch, test, and promote changes in a controlled path: personal → test → preview → release, ensuring production remains stable while iteration stays fast.
Last updated
Was this helpful?

