Glean Agents

Enterprise

Glean

Product overview

Name of Agent: Glean Agents
Short description of agent: "The horizontal agent environment for creating and managing AI agents at scale" (link, archived)
Date of release: 20/05/2025 (link, archived)
Advertised use: HR processes, e.g. performance reviews, research and bringing together various sources, product manager (link, archived)
Monetisation/Usage price: Unclear
Who is using it?: enterprises for data connection and management
Category: Enterprise

Company & accountability

Developer: Glean
Name of legal entity: Glean Technologies, Inc.
Place of legal incorporation: Delaware, USA (link, archived)
For profit company?: Yes
Parent company?: Not applicable
Governance documents analysis: Customer TOS (link, archived)
AI safety/trust framework: Responsible AI Policy (link, archived), Information Security Policy Technical Vulnerability Management Policy Data Management Policy Incident Response Policy (all locked).
Compliance with existing standards: SOC 2 ISO 27001:2022 HIPAA GDPR (link, archived)

Technical capabilities & system architecture

Model specifications: Page 12 of whitepaper (link, archived) suggests OpenAI, Anthropic, and Gemini frontier models are used.
Documention: No clear in depth documentation. Main page (here, archived).
Observation space: Has access to a number of tools through Glean's MCP (link, archived), and enterprise data through the enterprise graph (for more details on the enterprise graph, see (here, archived)).
Action space: Can edit enterprise information through enterprise graph, and make external tool calls (link, archived)
Memory architecture: Docs state "Complete long-term memory with the Enterprise Graph" (link, archived)
User interface and interaction design: Chatbot for users, node-based canvas for designers
User roles: Designer, Operator, Executor, Examiner
Component accessibility: Closed source, some elements are open sourced (link)

Autonomy & control

Autonomy level and planning depth: Designer is L1-L2: AI assistance is available in the editor, but user can make manual changes too. Resulting agent is L5: does all planning and execution, did not see any nodes for getting user approval/interaction
User approval requirements for different decision types: None, in the actual agent. User can respond to clarifying questions in the designer.
Execution monitoring, traces, and transparency: Pretty detailed CoT for both planning and execution. Allows pausing and replaying for debugging
Emergency stop and shut down mechanisms and user control: User can pause/stop the agent at any time
Usage monitoring and statistics and patterns: None found

Ecosystem interaction

Identify to humans?: None, it seems like the customer or developer designs the workflow, and Glean does not enforce AI outputs to be explicitly labelled as AI generated.
Identifies technically?: - Glean has its own crawling system as well (link, archived) - Glean's website crawler connector appears in third-party logs as GleanSearchCrawlerBot (link, archived) but could not find anything official in the Glean docs
Interoperability standards and integrations: - MCP support exists, and plans to support Agent Protocol (LangChain), Agent2Agent (Google), and AGNTCY (Cisco/LangChain/Galileo) (link, archived)
Web conduct: Glean’s Website connector (a crawler/indexer that can supply content to Glean Agents via Glean knowledge) states it “Respects robots.txt for crawl compliance” and includes an “Honor robots.txt” setting (enabled by default unless you have explicit approval to ignore it) (link, archived) For Glean Agents’ Web Search feature, Glean notes that for some providers it “does not crawl the content of the URLs” and also says it avoids paywalled sites that have robots.txt instructions not to fetch (link, archived)

Safety, evaluation & impact

Technical guardrails and safety measures: AI-based guardrails (link, archived) can be added to protect against prompt injection.
Sandboxing and containment approaches: None found
What types of risks were evaluated?: None found
(Internal) safety evaluations and results: None found
Third-party testing, audits, and red-teaming: None found
Benchmark performance and demonstrated capabilities: Glean reports a benchmark-style metric for its agentic system (Agentic Engine 2, used to power Glean’s agentic Assistant/Agents): 94% “task completeness” on end-to-end tasks, although it is unclear what exactly this measures (link, archived)
Bug bounty programmes and vulnerability disclosure: Yes (link, archived)
Any known incidents?: None found
← Back to 2025 Index