供资
为AI Agent身份提供700万美元的Minitive 融资，使资金总额达到1 100万美元
在 Google 上的首选来源中添加 Unite.AI
Minitive公司拥有由牛仔风险投资公司牵头的700万美元的融资新资金，随着公司寻求对其系统内运行的AI agents进行更严格的控制，其总资金达到1100万美元。 称公司现在每天处理250万个agent电话，超过50个支付企业客户，7个月内处理超过2亿个电话。
10月29日的公告名称为"客户"，包括"Ro","现代财政部","Coursera","密歇根大学"和"Apax". 资金支持一个平台，旨在给每个agent提供自己的身份，范围狭小的许可，以及一个可追踪的记录，记录其跨越企业工具和AI模式的行动。
为企业周围基础设施提供的资金Agents
Minitive的投资者包括Coatue、运营商集体、Burst Capital、Maven Ventures和Hustle基金。 个人支持者包括Okta联合创始人弗雷德里克·克雷斯特，WorkOS创始人兼CEO迈克尔·格里尼奇，斯科特·贝尔斯基，杰夫·迪恩和安德烈·卡纳德。 公告没有披露估价或指定新一轮谈判的阶段。
根本问题是许可管理。 雇员可能可以使用许多系统，因为他们的责任涉及多个任务。 agent使用该员工的资质可以继承比其即时工作需要的更大的范围。 随着各组织通过不同的接口部署agents，安保小组也面临零散的控制，对谁授权采取行动的说法不完整。
“Agents已经在公司系统内部工作，每个公司都需要自己的身份和许可，就像每个雇员一样。”
如何将 Agent 身份从用户证书中分离
该公司的Agent身份证明文件将每个agent描述为它网关内的一个单独的委托人。 它在审计日志中有自己的工具捆绑、状态和身份。 访问可由启动任务的用户、团队证书或为工作提供的服务账户授权。
基本服务证书留在网关。 agent通过自己的证书认证，并获得短寿命的代币，而管理员则控制它能够接触到的单个工具。 撤销agent的令牌可以阻止agent而不要求一个组织禁用无关的用户或agents.
这种分离还涉及连续性。 业务流程不应无限期地依赖于最初配置该流程的雇员的个人证书。 Minitive描述了agent捆绑的共享管理和可转让所有权，允许团队随着责任的改变管理访问。
Gateways 输入工具和模型调用
Mintive将用于工具访问的MCP Gateway与用于模型调用的LLM Gateway合并。 模式上下文协议提供了AI应用程序和外部工具之间的共同接口；治理问题在于，应当允许哪些特定的agent连接。
MCP 时速网关产品页面描述已核准的服务器的内部目录，归为特定角色的捆绑。 管理员可以禁用个人工具，包括破坏性操作，并在不同的AI客户端实施共同的接入政策。 网关还可以在企业认证后托管内部的MCP服务器。
每通电话都有用户、工具、目的地服务器和护栏记录结果。 该产品支持将日志输出到安全信息和事件管理系统，并通过OpenTeleometter的OTLP协议将其流出。 这使安全小组能够将agent活动纳入其现有的监测进程。
Mint Guard 新增筛选出提示词注入，暴露出秘密，个人数据等风险。 称公司的平台是SOC 2 Type II审计。 这些控制有助于界定允许的行为并使得活动可以检查；其有效性仍然取决于一个组织所配置的政策和一体化。
客户工作流量在实践中显示访问问题
在现代财政部，称公司在技术账户管理者干预前，银行业务agent搜索哨兵，Notion,Linear，源代码和内部工具。 Mintive报告说，较简单的案件节省了30分钟，复杂的案件节省了最多4小时。 这些是公告中报告的针对客户的结果。
密歇根大学使用Mintive作为企业MCP网关，在中央IT管理访问时为被批准的工具提供共享的路线。 Mintive还提供Slack和Teams的同工agents，将企业工具访问与长期内存和计划工作相结合。
正如Unite.AI对MCP连接层上方的企业软件的分析所解释的，使一个工具可以到达只是提供有用的工作流程的一部分。 Minitive正在处理与该连接相关的身份、权限和业务记录。
拥有AI和软件基础设施经验的创始团队
Mintive，曾被称作MintMCP，在其官方媒体资料袋中列出了四个共同创始人：吉泉·恩吉亚姆，维杰·瓦苏德万，博日·西和菲利普·本赛德。 发布会将Ngiam的背景描述为Coursera的创始工程师和Google Brain的高级员工软件工程师，而Vasudevan则在Google帮助创建了TensorFlow.
新的资金将公司置于企业AI周围日益扩大的运营层：决定agent可以做什么，它使用谁的权威，以及之后还有什么证据。 其报告的使用情况表明，这些问题已经是客户在整个业务系统部署agents的生产要求。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Funding
Mintive Raises $7 Million for AI Agent Identity, Bringing Total Funding to $11 Million
Add Unite.AI to your preferred sources on Google
Mintive has raised $7 million in new funding led by Cowboy Ventures, bringing its total financing to $11 million as companies seek tighter control over AI agents operating inside their systems. The company says it now handles 2.5 million agent calls a day across more than 50 paying enterprise customers, with over 200 million calls processed in seven months.
The October 29 announcement names customers including Ro, Modern Treasury, Coursera, the University of Michigan, and Apax. The funding backs a platform intended to give each agent its own identity, narrowly scoped permissions, and a traceable record of its actions across enterprise tools and AI models.
Funding for the Infrastructure Around Enterprise Agents
Mintive’s investors include Coatue, Operator Collective, Burst Capital, Maven Ventures, and Hustle Fund. Individual backers include Okta co-founder Frederic Kerrest, WorkOS founder and CEO Michael Grinich, Scott Belsky, Jeff Dean, and Andrej Karpathy. The announcement does not disclose a valuation or designate the new round’s stage.
The underlying problem is permission management. An employee may have access to many systems because their responsibilities span multiple tasks. An agent using that employee’s credentials can inherit a much larger reach than its immediate job requires. As organizations deploy agents through different interfaces, security teams also face fragmented controls and incomplete accounts of who authorized an action.
“Agents are already working inside company systems, and each one needs its own identity and permissions, just as every employee does,” said Jiquan “JQ” Ngiam, Mintive’s co-founder and CEO, in the announcement.
How Mintive Separates Agent Identity From User Credentials
The company’s Agent Identities documentation describes each agent as a separate principal within its gateway. It has its own tool bundle, status, and identity in the audit log. Access can be delegated from the user triggering a task, a team credential, or a service account provisioned for the work.
Underlying service credentials stay in the gateway. The agent authenticates through its own credential and obtains short-lived tokens, while administrators control which individual tools it can reach. Revoking an agent’s token can stop that agent without requiring an organization to disable unrelated users or agents.
This separation also addresses continuity. A business process should not depend indefinitely on the personal credentials of the employee who first configured it. Mintive describes shared administration and transferable ownership for agent bundles, allowing teams to manage access as responsibilities change.
Gateways Govern Both Tools and Model Calls
Mintive combines an MCP Gateway for tool access with an LLM Gateway for model calls. Model Context Protocol provides a common interface between AI applications and external tools; the governance question is which of those connections a particular agent should be allowed to use.
The MCP Gateway product page describes an internal catalog of approved servers, grouped into role-specific bundles. Administrators can disable individual tools, including destructive operations, and apply a common access policy across different AI clients. The gateway can also host internal MCP servers behind enterprise authentication.
Each call is logged with its user, tool, destination server, and guardrail result. The product supports exporting logs to security information and event management systems and streaming them through OpenTelemetry’s OTLP protocol. That gives security teams a way to bring agent activity into their existing monitoring processes.
Mint Guard adds screening for prompt injection, exposed secrets, personal data, and other risks. The company says its platform is SOC 2 Type II audited. These controls help define permitted behavior and make activity inspectable; their effectiveness still depends on the policies and integrations an organization configures.
Customer Workflows Show the Access Problem in Practice
At Modern Treasury, the company says a bank-operations agent searches Sentry, Notion, Linear, source code, and internal tools before a technical account manager intervenes. Mintive reports savings of 30 minutes on simpler cases and up to four hours on complex ones. Those are customer-specific results reported in the announcement.
The University of Michigan uses Mintive as an enterprise MCP gateway, providing a shared route to approved tools while central IT manages access. Mintive also offers Coworker agents in Slack and Teams, combining enterprise tool access with long-term memory and scheduled work.
As Unite.AI’s analysis of enterprise software above the MCP connection layer explains, making a tool reachable is only part of delivering a useful workflow. Mintive is addressing the identity, permissions, and operational record surrounding that connection.
A Founding Team With AI and Software Infrastructure Experience
Mintive, previously known as MintMCP, lists four co-founders in its official media kit: Jiquan Ngiam, Vijay Vasudevan, Bozhi See, and Phillip Bensaid. The release describes Ngiam’s background as Coursera’s founding engineer and a senior staff software engineer at Google Brain, while Vasudevan helped create TensorFlow at Google.
The new funding places the company in a growing operational layer around enterprise AI: deciding which agent can act, whose authority it uses, and what evidence remains afterward. Its reported usage suggests those questions are already production requirements for customers deploying agents across business systems.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.