Glean

Agents take authenticated action in outside systems, not only read from them. The surface is published as Connectors & Actions, and the orchestration page states agents "connect to external systems" so automations run where they are needed, with the security control point enforcing "source-system permissions on every read and write", which confirms that action, not only retrieval, is in scope.

The catalog is large and individually addressable: 275+ app connectors, each with its own page under glean.com/connectors, spanning documents, tickets, chat and code. The custom route is open too: the Indexing SDK "lets teams add custom, permission-aware data alongside Glean's existing connectors", and custom actions are documented on the APIs page. The same connectors also feed Glean's index.

Sourceglean.com/ai-agents connector index and glean.com/platform/apiread 2026-09-12

Multi step and multi agent execution are both documented. "Glean's agentic engine lets agents reason through tasks, plan next steps, and take action using enterprise context", which is plan and act rather than a fixed chain.

The orchestration surface adds work shared across agents: "route tasks between agents, and connect to external systems so your automations run exactly when and where they're needed", under a dedicated Agent orchestration page subtitled "Automate work across systems". That is agent to agent routing inside a governed runtime.

Supporting surfaces are an agent builder for "reasoning-based agents and complex workflows" and a deployment surface to "roll out, share, and certify agents".

Sourceglean.com/ai-agentsread 2026-09-12

A persistent index and graph sit over the customer's corpus. Glean states "275+ connectors across apps, documents, tickets, chat, and code" and "1000+ signals in the Enterprise Graph mapping people, content, and relationships", and its operating principle is "index your data once, then use it everywhere." The Enterprise Graph builds "not just content but the relationships between people, data, and activity across your organization."

That structure persists between runs and stays queryable, rather than being assembled per run. It is permissions aware at retrieval time, enforcing source system permissions on every read, so it carries access state as well as content.

Sourceglean.com/platform/enterprise-contextread 2026-09-12

A pre execution gate ships with the product through Glean Protect, rather than relying on one the customer must already own.

Two control points do the work: Action Alignment, to "check agent actions against the user's intent and authority before execution", and the headline commitment to "check write actions before they execute to keep agents aligned with their intended scope."

A third supports them: Topic Restrictions, to "check requests against acceptable use and industry-specific policies before the model responds." That gate inspects a proposed write and can stop it before it lands, enforced by Glean rather than by the downstream system.

On the same page, "enforce agent governance policies" and "different policies for select agents or users" are both marked coming soon, and "keep agentic browser use in check" is in beta. The two control points above carry no such marker. No approval queue that a person reviews item by item is documented.

Sourceglean.com/platform/securityread 2026-09-12

The Glean Protect page covers both compliance and access. SOC 2 Type II, ISO 27001, ISO 42001, HIPAA, TX-RAMP Level 2 and GDPR are displayed as held rather than in progress, with a trust center at trust.glean.com.

For access it names Identity & Access, to "verify who is making the request via your SSO provider before AI runs, including both user and agent identity", and Data Permissions, to "enforce source-system permissions on every read and write; an agent should not automatically access everything the user can."

That is identity integration through the customer's own provider plus a named permission model, extended to agent identity as a first class principal.

The page also carries an always on audit trail, which records what happened rather than controlling access. Four items on the page are marked coming soon or beta: agent governance policies, per agent and per user policies, Enterprise Graph enriched threat detection, and security scans of memory.

Sourceglean.com/platform/securityread 2026-09-12

Auditing is documented, not only reporting: "Always-on auditing. Every control point is logged to an end-to-end audit trail as it happens, with events available to stream into SIEM."

Six named control points are logged, identity and access, data permissions, sensitive content, topic restrictions, runtime threat protection and action alignment, so the trail records who asked, under what identity, what was permitted, what was screened and what the agent was allowed to do. That reconstructs why, not just what, and SIEM streaming makes it the buyer's record rather than a vendor dashboard. The trail is Glean's own record of what its agents did, not the logs the customer's Slack, Jira or Salesforce already keep.

A second, separate surface, agent observability, gives "deep insights into adoption, error rates, upvotes, downvotes, and ROI". Those are operational metrics, separate from the audit trail.

Sourceglean.com/platform/security and glean.com/ai-agentsread 2026-09-12

Memory is named as a component, but nothing about its scope, lifetime or surfaces is documented. The platform overview lists "Enterprise Memory learns continuously" as one of four Enterprise Context components, but what is published under that heading is the Enterprise Graph, a knowledge structure rather than agent memory. No per user or per tenant scope for the memory itself, published lifetime, expiry or purge path, or read or write surface a buyer could address is documented. A Personal Graph page sits at glean.com/enterprise-context/personal-graph.

Proactive AI, which "builds a live view of your teams and projects" and anticipates next steps, is marked coming soon on the vendor's own page.

Sourceglean.com/platform/enterprise-context and glean.com/platformread 2026-09-12

Customers can host Glean in their own cloud tenant. On its own Glean Protect page, Glean states that "SafetyCulture chose Glean in part for its security-first approach to enterprise data and the ability to host Glean within its own cloud tenant." TX-RAMP Level 2 points to a deployment posture US state governments can procure.

The statement comes from a customer story rather than a deployment page, and no single tenant, customer managed, on premises or air gapped deployment is published. A trust center sits at trust.glean.com.

Sourceglean.com/platform/securityread 2026-09-12

Customers can adopt agents from a browsable catalog. The Agent Library is a named product surface whose stated purpose is to "discover trusted, reusable agents", positioned so every team can "build, deploy, and discover new agents" on one platform, with a deployment surface to "roll out, share, and certify agents".

Discovery plus certification plus adoption is a catalog a buyer selects from, not a product line, and the library holds independent agents built for different jobs, so removing any one leaves the rest whole. No count of prebuilt agents is published. The agent builder is a construction surface rather than a packaged asset.

Sourceglean.com/ai-agents and its Agent Library sectionread 2026-09-12

Key events start agent runs: "Trigger agents from key events, route tasks between agents, and connect to external systems so your automations run exactly when and where they're needed." Event started runs are the published purpose of the orchestration surface. Channel coverage sits on top: Glean is reachable from its own web app, from 275+ connected applications, and through Glean MCP inside Claude Code, Codex, Gemini, Cursor and Copilot.

Proactive AI, which "anticipates next steps like meeting prep or first drafts before you ask", is marked coming soon on Glean's Enterprise Context page; the event triggers carry no such marker.

Sourceglean.com/ai-agents, against glean.com/platform/enterprise-contextread 2026-09-12

Glean documents both vendor routing and customer control of models. "Glean Intelligence optimizes AI for price-performance, matching each task to the right model across 40+ open and frontier models", with a dedicated Auto routing page.

Customer control ships as named surfaces: a Model Hub whose stated purpose is to "get access to the latest models", "flexible model choice" listed as a property of the Open Platform, an AI Gateway to "govern every AI interaction", and Usage Controls to "control AI usage and spend." A hub the admin picks from, plus a gateway that governs which models may be used, is customer and admin control of the model.

Multi cloud support and the Bedrock, Vertex and Azure OpenAI hosting relationships are infrastructure, not separate model providers. No page enumerates the per step selection interface.

Sourceglean.com/platformread 2026-09-12

Outside callers reach Glean through documented APIs. Named interfaces for its own platform are the Search API, Chat API and Agents API, with an Indexing SDK that "lets teams add custom, permission-aware data alongside Glean's existing connectors" and a Web SDK to "embed Glean UI and context-aware chat or agent experiences". The Agents API "follows the LangChain Agent Protocol", with "client libraries available in TypeScript, Go, Java, and Python." An ungated developer portal sits at developers.glean.com, linked from the site footer.

Glean MCP "feeds governed enterprise context into Claude Code, Codex, Gemini, Cursor, Copilot", an outside caller driving Glean. The 275+ connectors run the other way, letting Glean drive other systems.

Sourceglean.com/platform/api and glean.com/blog/go-agents-launchread 2026-09-12

Glean asserts a testing step but documents no evaluation harness, scored test case, quality gate or controlled experiment. The agent builder, at glean.com/ai-agents/agent-builder, is headed "Build, test, and iterate with ease", which asserts a test step without describing one: no scoring, comparison, pass criterion or result a buyer reads. The deployment surface offers to "roll out, share, and certify agents", but certification reads as an administrative blessing for discovery rather than a gate on agent behavior, and nothing states what certifying tests.

Agent observability supplies "deep insights into adoption, error rates, upvotes, downvotes, and ROI", but those are live operational metrics rather than tests, and no scored runs against test cases are published.

Sourceglean.com/ai-agentsread 2026-09-12

Computer use is not documented: no hosted browser, desktop session or remote or local computer control appears. Glean reaches other systems through 275+ API connectors and custom actions, which are programmatic interfaces rather than interface operation, so breakage when UI elements change does not arise.

Glean Protect lists "keep agentic browser use in check (beta)" among its agent guardrails. That is a governance control over browser use, and on a platform whose stated role includes feeding context into Claude Code, Codex, Cursor and Copilot, the browser use being governed is most plausibly other tools' rather than Glean's own. It is in beta, and a control over a capability does not show the vendor ships it; no documentation shows Glean's own agent driving a browser.

Sourceglean.com/platform/security and glean.com/platformread 2026-09-12