Feature availability by LLM key, provider, and cloud
This page documents which model providers and which Glean Assistant features are available depending on:
- Key type — the Glean Universal Model Key (Glean-managed) or Customer Key (BYOK), where you bring your own provider key.
- Cloud environment — Glean Hosted or Customer Hosted on GCP or AWS.
- LLM provider — OpenAI, Azure OpenAI, Google Vertex AI (Gemini and Claude), or Amazon Bedrock (Claude and Amazon Nova).
Use these matrices to understand what changes when you move from the Glean Universal Model Key to your own key, and to plan which models you need to enable for a given feature.
note
With the Glean Universal Model Key, Glean manages connectivity to all supported providers regardless of your cloud environment. With Customer Key (BYOK), provider access is limited to cloud-native providers because cross-cloud access is not supported.
1. Provider availability by key type and cloud
Which LLM hosting providers you can configure depends on your deployment mode and key type.
| Hosting provider | Glean Universal Model Key | Customer Key — GCP-based deployment | Customer Key — AWS-based deployment |
|---|---|---|---|
| OpenAI | ✅ | ✅ | ✅ |
| Azure OpenAI | ✅ | ✅ | ✅ |
| Google Vertex AI (Gemini, Claude) | ✅ | ✅ | ❌ |
| Amazon Bedrock (Claude, Amazon Nova) | ✅ | ❌ | ✅ |
note
- GCP-based deployment includes Glean Hosted and Customer Hosted on GCP. AWS-based deployment refers to Customer Hosted on AWS.
- Anthropic (Claude) models are accessible through either Google Vertex AI or Amazon Bedrock, depending on your available providers.
- With Customer Key (BYOK), provider access is limited to cloud-native providers because cross-cloud access is not supported. GCP-based deployments cannot use AWS-hosted models (for example, Amazon Bedrock), and AWS-based deployments cannot use Vertex AI.
- For Customer Key deployments, the large, small, agentic, and fast agentic model defaults must come from the same provider. Image generation models may use a different provider.
2. Feature availability by key type
Status legend: GA = generally available · Beta = limited/early access · Not supported = not available on that key type today.
"Required model(s)" lists the model(s) or provider(s) a feature depends on. Where a feature has no specific model dependency, it uses your standard configured Assistant models.
| Feature | Glean Universal Model Key | Customer Key (BYOK) | Required model(s) / provider | Notes |
|---|---|---|---|---|
| Real-time voice | GA | Not supported | GPT Realtime 1.5 / GPT 4o Transcribe (OpenAI or Azure OpenAI) | Customer Key support is not yet available. Audio is routed through Glean to OpenAI; no raw audio is stored. The admin setting is only visible for Glean Universal Model Key customers. |
| Meeting Notes | GA | Not supported | Standard Assistant models | Desktop, Mac-only. Generally available to Glean Key customers on the Glean macOS desktop app. Support for customer-key and self-hosted deployments is not yet available. |
| Deep Research | GA | Not supported | GPT 5.1 (reasoning); Brave Web Search for web content | Per the current Deep Research documentation, it works exclusively with the Glean Universal Key and is not compatible with legacy keys (Azure, Vertex, AWS) or customer-hosted keys. Web application only. |
| Image generation | GA | GA (all clouds) | GPT Image 2 / GPT Image 1.5 (OpenAI or Azure); Nano Banana Pro (Gemini Pro Image 3) / Gemini Flash Image (Vertex AI) | Available for Customer Key on all clouds. Requires enabling an image-capable model. Image generation models can use a different provider than your default models. Claude models do not generate images. |
| Code generation (Code Writer) | GA | Conditional | Amazon Bedrock Claude in the same project (BYOK) | For Glean-billed customers, Code Writer uses the same Vertex AI configuration as Glean. For bring-your-own-key deployments, Code Writer works only with a compatible Amazon Bedrock instance in the same project; third-party Anthropic keys are not used. |
| Slide deck generation | GA | GA | Standard Assistant models | PowerPoint and Google Slides export. Off by default; an admin must upload a company .pptx template and enable the feature. |
| Spreadsheet generation | GA | GA | Standard Assistant models | Available wherever Glean and Canvas are enabled. No separate template or admin setup. |
| Interactive pages (HTML artifacts) | GA | GA | Standard Assistant models | Generate dashboards, infographics, and calculators as HTML, previewed in Canvas. |
| Memory & personalization | GA (GCP) | GA (GCP) | Standard Assistant models | Available for GCP deployments only. AWS self-hosted deployments are not supported, with no admin or tenant-side control to enable it. |
| Web search | GA | GA | Web provider (for example, Google Enterprise Search / Brave); standard Assistant models | Used in All Knowledge mode to blend real-time web information with company data. |
| File upload & analysis | GA | GA | Standard Assistant models (upload limits vary by model) | Upload and analyze local files, including .zip analysis in Thinking mode. Video and audio files are not supported. |
| Contextual images in responses | GA | GA (GCP, AWS) | Standard Assistant models | Displays relevant images from Google Drive and SharePoint documents. Enabled by default for most Glean hosted (GCP) customers; self-hosted requires GCP or AWS. |
| Model choice in Assistant | GA | GA | Models the admin has enabled | Customer Key users can select from a model-choice dropdown of the models their organization has enabled. |
| Agent Sandbox & Programmatic Tool Calling (PTC) | GA (Glean-hosted) | GA (Customer-hosted on GCP, AWS) | Frontier model recommended: GPT 5.4+, Claude Sonnet 4.6, Claude Opus 4.6+, or Gemini 3.1 Pro | Runs code-backed analysis in a secure sandbox in Thinking mode. PTC only runs inside an active sandbox session. |
| Agentic Search Model | GA (Glean-hosted, GCP) | Not supported | Glean retrieval-optimized model | Available only to customers hosted by Glean on GCP using the Glean Universal Model Key. Not applicable to AWS, Azure, customer-managed key, or EU deployments. |
| LLM / Model Insights (BYOK) | Not supported | GA | N/A | Customer Key admins can monitor Glean-attributed LLM usage and reliability (peak/avg TPM, peak RPM, 429 rate, 5xx rate, with per-model breakdowns). Available only for Customer Key (BYOK) deployments. |
| Best-fit default model selection | GA | Not applicable | Glean-managed model selection | With the Glean Universal Model Key, Glean automatically uses best-in-class basic and standard models by default. On Customer Key, admins choose the default models. |
3. Agentic engine mapping to keys and providers
The agentic engine that powers Glean Assistant runs on GPT-5 or Claude Sonnet 4.6:
- GPT-5 is available via Glean Key or Customer Key through OpenAI Direct or Azure OpenAI.
- Claude Sonnet 4.6 is available via Customer Key through Google Vertex AI or Amazon Bedrock.
- Glean Key customers are automatically migrated to GPT-5. Customer Key customers must manually migrate their models to GPT-5 or Claude Sonnet 4.6 to see a performance difference.
note
For Glean Assistant on Customer Key, Gemini Pro 3.1 is not available — use Gemini Pro Custom Tools 3.1 instead. Gemini Pro 3.1 remains available for Glean Assistant through the Glean Universal Model Key.
4. Model upgrades, failover, and deprecation
- Glean Universal Model Key: Glean manages model configuration, provisioning, upgrades, and provider failover.
- Customer Key (BYOK): Your team provisions capacity and manages model upgrades and provider failover. Quota is not a guarantee of capacity; for guaranteed capacity, contact your provider about provisioned throughput.
- When a model is deprecated, existing deployments are automatically migrated to the listed replacement. Review the Model Hub deprecation table to confirm your configured models are current.
See also
- Set up LLMs using the Model Hub — provider/model configuration and the availability matrix.
- Real-time voice — setup and limitations.
- Image generation in Glean — supported image models and setup.
- Memory and personalization — admin configuration and known limitations.
- Agent Sandbox & Programmatic Tool Calling — setup and supported models.
- LLM Insights — monitoring for Customer Key (BYOK) deployments.