Glean

Glean crawls the applications a company already runs and builds a single index with the source system's access controls attached to every document. When a query arrives, the signed-in user's identity is evaluated against those mirrored permissions before anything is returned, and the same check applies to search results, generated answers and their citations, agent runs, MCP calls and the tiles embedded in ServiceNow, Zendesk and Salesforce. Glean states that if a user can't open a document in the source system, it cannot be used as context or surfaced as a citation.

The catalog runs to 275+ out-of-the-box connectors in three kinds: native indexing, Push API and MCP-based. On top of that index sit Glean Search, Glean Assistant (chat, document upload, data analysis, canvas, slide and image generation, deep research, meeting notes) and Glean Agents.

Agent Builder offers two routes. A user can describe the agent's purpose in plain language and have steps generated, or assemble steps by hand, switching between the two at any point. Which tools appear depends on what an admin has enabled and whose access scope the user falls in, so the builder is open to business users while the tool surface stays administered.

Autonomy is deliberately incomplete. Write actions running in the Glean web app pause for user approval with a panel showing the target application and the planned change. Administrators can mark a specific action as safe to run without that confirmation, which Glean describes as an explicit opt-in rather than the default, and the confirmation does not apply in Slack or Microsoft Teams. Agents with scheduled triggers run in the background, where control shifts to who may publish agents and which actions are enabled for autonomous use.

Workflow mode agents export to JSON and import into another deployment, with tool references made portable and owner names obfuscated. Auto mode export is a ZIP of spec.yaml and instructions.md, recorded as limited availability pending a wider rollout.

There is no pricing page on glean.com and no pricing URL among the 1,883 entries in its sitemap. What exists instead are two documentation pages that describe the licensing in detail. Glean Core Suite is licensed per user per month and includes the enterprise graph, connectors, Glean Protect, the admin surfaces, the platform and admin APIs and the indexing APIs. Anything that calls a model sits outside the seat under Model Hub Usage, bought as an upfront commit and charged at provider API rates, published per million tokens down to each individual model.

Glean Enterprise Flex is the newer shape: a per-user seat plus a pooled allowance of FlexCredits. Fast Mode queries, collections, go links, the people directory and agent creation are unlimited. Thinking Mode and Adaptive Reasoning queries on standard models are included up to 100 per user per week, after which they draw credits. Premium models, Code Writer, slide generation, meeting notes, voice sessions, learned memories and every agent run draw credits from the start. Credit consumption is listed as Variable with median and 90th percentile typicals: an advanced agent run is roughly 150 credits at the median, roughly 450 at the 90th.

Two items are sold as separate site-wide annual fees: Protect+, which adds sensitive content models and prompt injection protection, and Premium Support, which carries the only written response commitment Glean publishes. Credits and seats are discounted for customers who bring their own model keys or self-host.

The Trust Center names SOC 2, ISO 27001:2022, ISO/IEC 42001:2023, HIPAA and GDPR, and lists the corresponding report and certificate documents, along with a control stating that customer data is deleted upon leaving. The documents themselves require an access request, and Glean's own docs say compliance reports and certifications are covered by confidentiality agreements. Glean Technologies is certified under the EU-U.S. Data Privacy Framework and its UK and Swiss extensions, per its privacy statement, which covers the website and business operations rather than the product.

SSO is mandatory and is the first thing configured in a new tenant, over OIDC or SAML, with Glean recommending OIDC for finer permission control. Admin audit logs cover connector, role and configuration changes with filtering, CSV export and optional SIEM streaming; they retain 30 days by default and a longer window has to be arranged with an account manager. End-user activity lives in separate Customer Event logs, whose ongoing delivery is not self-serve.

Several Glean Protect capabilities carry qualifiers on the marketing page: agent governance policies, context-aware threat detections and memory scanning are marked coming soon, and security chat, agent scanning and agentic browser controls beta.

Glean can run inside the customer's own GCP or AWS account, with published region and service requirements, support for read-only CSPM tooling and Wiz or CrowdStrike runtime sensors, and an authoritative CVE report for Glean-built container images shipped with each release into the deployment's own release-notes bucket. GovCloud regions are not supported. Glean-hosted deployments are single-tenant with regional data residency included in the seat fee.

Model choice is administered through the Model Hub, with models from OpenAI, Google, Anthropic, Amazon and several open providers plus Glean's own Waldo model. A customer can use Glean's Universal Model Key or bring its own provider keys, in which case cross-cloud access is restricted: a GCP-based deployment cannot reach Amazon Bedrock and an AWS-based one cannot reach Vertex AI, with Anthropic the exception because it is reached directly. Premium models can be disabled entirely from the admin console, which is also the lever for keeping credit consumption down.

This is an administered platform, not something a team adopts on a Tuesday afternoon. There is no free plan, no trial, no sign-up and no published seat price; the entry point is a demo request, and deployment starts with an identity provider integration and connector credentials. A small company without an IT function has nothing to start from here.

Cost predictability is the second caution. Seats cover search and fast chat, and a meaningful share of what the marketing pages demonstrate, which is agent runs, deep research, slide generation and premium models, is metered in credits whose per-unit rate is published as Variable. Model Hub usage is a separate commitment charged at provider token rates. Glean publishes usage controls for setting monthly limits per organization, department, user and agent, which is an admission that the meter matters.

One figure sits oddly against the rest. The home page advertises 99.9%+ uptime across core search and AI services as a statistic, while the only written service commitment in the pricing documents is a one-hour critical issue response time inside Premium Support. No uptime SLA is published.