Deep Research
Turn Deep Research on or off, and check how a run is billed on each key. Users run Deep Research for multi-source reports such as planning briefs and competitive reviews. For the steps they follow, see Deep Research in the Skills user guide.
Check availability
Availability follows the key on the deployment:
- Glean Universal Model Key — Deep Research is a skill in chat. The per-user monthly quota doesn't block a skill run.
- Customer Key — Deep Research stays a chat mode, and the monthly quota stays in place. The skill itself has no separate quota.
Wherever Deep Research runs as a skill, users also need Skills access. See Require Skills access.
You turn Deep Research on or off on both keys. On a Customer Key, Assistant uses the model you set in the Model Hub. When Deep Research is a skill, it uses that same model. See Configure LLMs in the Model Hub.
Turn Deep Research on or off
You can turn Deep Research on or off for your organization, or limit it to specific users and identity provider (IdP) groups in the Admin console:
- Navigate to Admin console → Platform → Assistant → Settings.
- Expand the Deep Research section.
- Select one of the following:
- On for everyone
- On for selected teammates and groups (allowlist of named users, IdP groups, or both)
- Off
- If you chose On for selected teammates and groups, add entries in Users, Groups, or both. Add at least one entry.
- Select Save.
Changes take up to 30 minutes to take effect.
Set feature restrictions when you want a security-reviewed pilot group rather than a full rollout. For limits, sync behavior, and the other features that support allowlists, see Restrict feature access by user or group.
On the Glean Universal Model Key, users start a report from the skill picker or by typing /deep-research. The skill is available in the web app. On a Customer Key, Deep research still appears in the chat mode picker.
The monthly quota doesn't apply for Glean Universal Model Key organizations. It remains for Customer Key organizations that are still on the chat mode.
Require Skills access
Deep Research is a skill users start from chat, so it follows Skills in Assistant as well as the Deep Research setting. Turning Deep Research on doesn't grant Skills access.
A user can start a report only when both of these are true:
- Deep Research is On for everyone, or On for selected teammates and groups and the user is named or belongs to a selected group.
- Skills in Assistant is Enabled for all teammates, or Enabled for test group only and the user is in that test group.
If Skills is off for that user, Deep Research is missing from the skill picker and /deep-research doesn't start a report. Turn Skills on for anyone who should keep Deep Research. See Set up Skills and manage access.
On a Customer Key where Deep Research is still a chat mode, Deep research stays in the mode picker and Skills access isn't required. The rule applies wherever Deep Research runs as a skill.
Understand a Deep Research run
- Sources — searches connected enterprise systems and the web, then writes a report with linked citations. Web content comes from Brave Web Search.
- Model — a skill run uses the model selected for that chat. On the Glean Universal Model Key, that is the model Glean routes to.
- Permissions — each user only sees documents they can already open.
- Read-only — Deep Research does not run tools that change connected apps. Ask in chat when the task has to write back to an app.
- Length and time — a typical report is 5–10 pages and takes 5–30 minutes, depending on the question and the number of sources.
Review key support
Deep Research supports the Glean Universal Model Key and Customer Key (BYOK). A skill run uses the Assistant models already configured for chat, on the providers those models use: Azure OpenAI, Amazon Bedrock, and Google Vertex AI. See Bring Your Own Key (BYOK). See Feature availability by LLM key, provider, and cloud.
Review usage and billing
On the Glean Universal Model Key, a Deep Research run is billed as a normal Assistant query. On Glean Enterprise Flex, standard models follow your Assistant terms, and premium models consume FlexCredits. The separate monthly quota does not apply.
On a Customer Key, while Deep Research is still a chat mode, each report consumes FlexCredits, and the monthly quota still applies.