Central AI administration
Synology AI Console is the control plane for AI features used by supported DSM packages. It can connect approved packages to third-party AI APIs, manage user and group permissions, track integration status and usage, apply de-identification, and on GPU-powered systems manage local models for embedding, OCR, speech-to-text and image captioning. That makes it both a productivity feature and a governance boundary.
Quick answer
Configure governance before adding providers
Decide which DSM packages may use AI, which user groups are excluded, whether data de-identification is required and which provider/API endpoint is allowed before enabling broad access. For local GPU models, budget memory by model and leave system headroom instead of turning every model on by default.
Live Amazon products
Synology systems relevant to this AI and DSM decision
Product cards are discovery aids only. They never override Synology's live applied-model lists, GPU compatibility list, DSM release notes or package requirements.
Technical decision
Separate API integration from local model control
AI Console has two distinct roles. API integrations connect supported packages to external or OpenAI-compatible AI services under administrator-controlled permissions. GPU model control supports local embedding, OCR, STT and image-captioning features used by Drive AI Search. Treat those as separate data-flow and capacity decisions.
Interactive AI / DSM tool
AI Console Deployment Planner
Use this as a planning aid. It applies the inputs you provide, but final model support, GPU compatibility, package requirements, beta status and DSM behavior must be checked against Synology's current official documentation before purchase or production changes.
Deployment checklist
Four checks before a Synology AI rollout
Match the exact model
Do not infer support from a nearby model number, generation or shared CPU. Use Synology current current applied-model list.
Verify the DSM build
DSM Agent requires DSM 7.4 or later, while AI Search also depends on newer package versions and semantic indexing support.
Separate AI feature paths
DSM Agent, website AI Advisor, AI Console and Drive AI Search have different hardware, privacy and deployment requirements.
Pilot before rollout
Measure memory, indexing time, permissions and recovery implications on representative data before enabling AI features broadly.
AI Console is an administrator control plane
The package centralizes AI configuration for supported DSM applications. Administrators can define integrations, monitor status and usage, assign package access and restrict groups or users rather than configuring AI independently in every application. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on ai console is an administrator control plane adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
Supported API providers are broad
Synology lists several major AI platforms and compatible endpoints. The current specification includes Amazon Bedrock, Azure OpenAI, Baidu AI Cloud, Google AI Studio, Google Vertex AI, OpenAI and other services exposing OpenAI-compatible APIs. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on supported api providers are broad adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
Permissions should precede experimentation
Package access and user access are separate governance choices. Enable only the application that needs AI, then exclude groups or users who should not send prompts or content through the configured integration. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on permissions should precede experimentation adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
De-identification is a privacy control, not magic
AI Console can replace sensitive entities before cloud transmission. The feature requires Container Manager and at least 8 GB of memory, and Synology notes that automated masking cannot guarantee one hundred percent removal of sensitive information. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on de-identification is a privacy control, not magic adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
Language packs consume memory
De-identification can load language-specific models. Synology indicates each selected language pack uses roughly 1 GB, so multilingual deployments should plan memory rather than enabling every language by default. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on language packs consume memory adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
GPU model control powers local search features
AI Console also manages on-device models on supported GPU systems. Embedding supports semantic search, OCR extracts text, STT converts speech and image captioning generates searchable descriptions. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on gpu model control powers local search features adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
Memory budgeting is workload-specific
The local model set can consume several gigabytes before normal DSM services are counted. Synology lists about 2 GB each for embedding and OCR, and about 3 GB each for STT and image captioning, with a recommendation to reserve substantial system memory. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on memory budgeting is workload-specific adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
AI Search can optionally use an external LLM
Drive can combine local semantic indexing with API-based intent interpretation. When enabled, the LLM can interpret clear query intent and apply smart filters such as owner, file type or time range. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on ai search can optionally use an external llm adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
Provider choice affects data flow
A public-cloud API changes where request data is processed. Document the provider region, account ownership, retention terms, authentication method and the purpose of each integration before giving users access. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on provider choice affects data flow adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
OpenAI-compatible endpoints create flexibility
The compatible-API option can support services beyond the named provider list. That flexibility increases the importance of validating endpoint behavior, TLS, authentication, logging and contractual data handling rather than assuming compatibility equals security. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on openai-compatible endpoints create flexibility adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
Usage monitoring should become routine
AI costs and data flows can expand quietly. Review API consumption, package access and enabled groups on a schedule, then remove abandoned integrations and credentials. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on usage monitoring should become routine adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
A minimal configuration is easier to audit
Start with one package, one provider and a small user group. Once permissions, privacy behavior and cost monitoring are proven, expand intentionally instead of enabling AI across the NAS in one step. AI Console is primarily a governance surface, so configuration should begin with data flow and permissions. Name the provider, application, approved users, privacy controls and monitoring owner before expanding an integration, because convenience can otherwise hide a broad new path for organizational data.
The section on a minimal configuration is easier to audit adds a separate planning question. For this AI Console topic, separate cloud API integration from local GPU model control. Each path consumes different resources and creates different privacy obligations, which makes a small, auditable pilot easier to secure and troubleshoot than an all-at-once rollout.
Questions people ask
Synology AI Console questions
What does Synology AI Console do?
It centrally manages AI API integrations, package and user permissions, usage, privacy controls and local GPU AI models on supported systems.
Which AI providers are supported?
Synology currently lists Amazon Bedrock, Azure OpenAI, Baidu AI Cloud, Google AI Studio, Google Vertex AI, OpenAI and OpenAI-compatible APIs.
Does AI Console require DSM 7.4?
The AI Console technical specification supports DSM 7.2 or above, while newer Drive AI Search workflows depend on later package versions and supported GPU hardware.
What is de-identification?
It replaces selected sensitive entities with placeholders before data is sent to a configured public-cloud AI provider, then restores the original information after the response returns.
Is de-identification guaranteed to remove all sensitive data?
No. Synology explicitly notes that model- or rule-based de-identification cannot guarantee one hundred percent masking.
What does de-identification require?
Synology requires Container Manager and at least 8 GB of system memory for the feature.
Can AI Console run local AI models?
On supported GPU-powered storage systems, it can control embedding, OCR, speech-to-text and image-captioning models used by AI Search.
How much RAM do the local models use?
Synology lists roughly 2 GB for embedding, 2 GB for OCR, 3 GB for STT and 3 GB for image captioning, plus normal system requirements.
Can I exclude users from AI access?
Yes. AI Console includes permissions to enable package access while excluding specific groups or users.
Should I use one API key for everything?
A more auditable design is to separate integrations by purpose or environment when the provider allows it, then apply the least privilege and monitor usage.
Official references and methodology
Verify the exact model, DSM build, package versions and support state
Feature and memory details are based on the DSM 7.4 AI Console technical specification and Synology’s AI Search setup documentation. Privacy guidance treats de-identification as a risk-reduction feature rather than a guarantee.
As an Amazon Associate, Cloudzat may earn from qualifying purchases. Prices, product availability, DSM builds, beta eligibility, package versions, GPU compatibility and AI feature requirements can change. Verify the exact Synology model and current official release notes before purchasing hardware or changing a production system.