https://github.com/activeloopai/activeloop-self-hosted-resources

Official Helm charts for deploying Activeloop services on Kubernetes
https://github.com/activeloopai/activeloop-self-hosted-resources

Last synced: about 1 month ago
JSON representation

Official Helm charts for deploying Activeloop services on Kubernetes

Host: GitHub
URL: https://github.com/activeloopai/activeloop-self-hosted-resources
Owner: activeloopai
Created: 2025-07-10T07:50:45.000Z (8 months ago)
Default Branch: main
Last Pushed: 2025-12-25T07:43:49.000Z (3 months ago)
Last Synced: 2025-12-26T20:53:07.720Z (3 months ago)
Language: Smarty
Size: 10.1 MB
Stars: 0
Watchers: 0
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

# activeloop-self-hosted-resources

## Available Deployments

### Activeloop Neohorizon

- [K8s](./helm/activeloop-neohorizon/)
- [docker-compose](./docker-compose/activeloop-neohorizon/)

#### Parameters

| Helm Parameter | Environment Variable | Default Value | Descriptoin |
| ----------------------------------------------- | ----------------------------------------------- | ------------------------------------------------------------ | --------------------------------------------------------------------------------- |
| deeplake_creds | DEEPLAKE_CREDS | - | Refer to [Deeplake storage credentials](./README.md#deeplake-storage-credentials) |
| deeplake_root_dir | DEEPLAKE_ROOT_DIR | for `helm` required, for docker-compose: `/var/lib/deeplake` | storage path used by deeplake for data operations |
| postgres_database | POSTGRES_DATABASE | neohorizon | virtual database name to use by app, must be created beforehand |
| postgres_host | POSTGRES_HOST | k8s service host of postgres dependency | postgres database hostname (required) |
| postgres_password | POSTGRES_PASSWORD | postgres | postgres database user password |
| postgres_user | POSTGRES_USER | postgres | postgres database username |
| postgres_port | POSTGRES_PORT | 5432 | postgres database port |
| rabbitmq_url | RABBITMQ_URL | - | rabbitmq ampq url, default will be built from dependency installation (required) |
| al_api_token | AL_API_TOKEN | - (required) | api token to authenticate to deployed api |
| gemini_api_key | GEMINI_API_KEY | - | optional to run geminy requests |
| openai_api_key | OPENAI_API_KEY | - | needed for query generation |
| text_image__matrix_of_embeddings__ingestion_url | TEXT_IMAGE__MATRIX_OF_EMBEDDINGS__INGESTION_URL | - | should be full path to endpoint triton inference endpoint |
| text_image__matrix_of_embeddings__query_url | TEXT_IMAGE__MATRIX_OF_EMBEDDINGS__QUERY_URL | - | should be full path to endpoint triton inference endpoint |
| text_image__embedding__ingestion_url | TEXT_IMAGE__EMBEDDING__INGESTION_URL | - | should be full path to endpoint triton inference endpoint |
| text_image__embedding__query_url | TEXT_IMAGE__EMBEDDING__QUERY_URL | - | should be full path to endpoint triton inference endpoint |
| text__embedding__ingestion_url | TEXT__EMBEDDING__INGESTION_URL | - | should be full path to endpoint triton inference endpoint |
| text__embedding__query_url | TEXT__EMBEDDING__QUERY_URL | - | should be full path to endpoint triton inference endpoint |

#### Models usage

Neohorizon works with triton served models for embedding generation both for queries and ingestion. Both helm chart and docker-compose are providing options to run models.
Here are descriptions of models:

##### Models that [Activeloop](chat.activeloop.ai) uses

- **colnomic**: can be used for ingest/retrieval of images, suggested to provide at least **16GiB** RAM and **A100** GPU
- **inf-retriever-v1**: can be used for ingest/retrieval of texts, suggested to provide at least **4GiB** RAM and **A10/L4** GPU
- **doclayout_parser**: can be used to generate images for answers, suggested to provide at least **4GiB** RAM and **A10/L4** GPU

##### Additional models we provide

- **qwen_06B**: can be used for ingest/retrieval, suggested to provide at least **4GiB** RAM and **A10/L4** GPU

> Note. Any custom models can be used with neohorizon, only requirement is that models must be served with triton and full URLs must be set in deployment environment variables.

For both helm chart and docker-compose cases default configuration should be reviewed or adjusted to use models.

- case 1: deployed only image model, then values override yaml would look like

- case 2: deployed all models with single deployment

```yaml
...
global:
config:
text_image__matrix_of_embeddings__ingestion_url: http://activeloop-neohorizon-models-svc/v2/models/colnomic/infer
text_image__matrix_of_embeddings__query_url: http://activeloop-neohorizon-models-svc/v2/models/colnomic/infer
text_image__embedding__ingestion_url: http://activeloop-neohorizon-models-svc/v2/models/colnomic/infer
text_image__embedding__query_url: http://activeloop-neohorizon-models-svc/v2/models/colnomic/infer
text__embedding__ingestion_url: http://activeloop-neohorizon-models-svc/v2/models/inf-retriever-v1/infer
text__embedding__query_url: http://activeloop-neohorizon-models-svc/v2/models/inf-retriever-v1/infer
...
models:
- name: models
load_models:
- colnomic
- inf-retriever-v1
```

- case 3: separate deployments for image and text

```yaml
...
global:
config:
text_image__matrix_of_embeddings__ingestion_url: http://activeloop-neohorizon-colnomic-svc/v2/models/colnomic/infer
text_image__matrix_of_embeddings__query_url: http://activeloop-neohorizon-colnomic-svc/v2/models/colnomic/infer
text_image__embedding__ingestion_url: http://activeloop-neohorizon-colnomic-svc/v2/models/colnomic/infer
text_image__embedding__query_url: http://activeloop-neohorizon-colnomic-svc/v2/models/colnomic/infer
text__embedding__ingestion_url: http://activeloop-neohorizon-text-svc/v2/models/inf-retriever-v1/infer
text__embedding__query_url: http://activeloop-neohorizon-text-svc/v2/models/inf-retriever-v1/infer
...
models:
- name: colnomic
load_models:
- colnomic
...
- name: text
load_models:
- inf-retriever-v1
```

#### Deeplake storage credentials

In the case cloud storage is used (s3, gs, azure blob storage) and underlaying infrastructure does not provide out of the box authentication to the storage,
static credentials should be applied as enviornment variable so deeplake can do storge operations.
To give credentials to deeplake, use `DEEPLAKE_CREDS` environment variable or corresponding Cloud SKD Environment variables.
`DEEPLAKE_CREDS` must be an string serialized dictionary with cloud credentials, examples below.

- **AWS**:

```jsonc
{
"aws_access_key_id": "AWS_ACCESS_KEY_ID",
"aws_secret_access_key": "AWS_SECRET_ACCESS_KEY",
"aws_session_token": "AWS_SESSION_TOKEN", // Optional
"endpoint_url": "https://s3.customerendpoint.com", // OPTIONAL
"region_name": "AWS_REGION" // OPTIONAL
}
```

```jsonc
{
"profile_name": "AWS_PROFILE",
"endpoint_url": "https://s3.customerendpoint.com", // OPTIONAL
"region_name": "AWS_REGION" // OPTIONAL
}
```

```jsonc
{
"aws_role_arn": "AWS_ROLE_ARN",
"aws_session_name": "session-name-for-assume-role",
"aws_external_id": "external-id-for-assume-role",
"endpoint_url": "https://s3.customerendpoint.com", // OPTIONAL
"region_name": "AWS_REGION" // OPTIONAL
}

- **AZURE**:

```jsonc
{
"azure_client_id": "AZURE_CLIENT_ID",
"azure_client_secret": "AZURE_CLIENT_SECRET",
"azure_tenant_id": "AZURE_TENANT"
}
```

```jsonc
{
"sas_token": "azure-storage-sas-token"
}
```

```jsonc
{
"account_name": "...",
"container_name": "...",
"account_key": "...",
}
```

- **GCP**:

```jsonc
{
"json_credentials": "SERVICE_ACCOUNT_JSON_KEY"
}
```

ecosyste.ms

Data

Tools

Indexes

Applications

Experiments

Awesome

https://github.com/activeloopai/activeloop-self-hosted-resources

Awesome Lists containing this project

README