Kindo Raises $30M to Expand Enterprise AI Agent Control

A person has to approve Kindo's event-filtering threshold before the software starts withholding work from an AI agent. That decision puts a human owner inside the economics of automation: somebody decides which events deserve a model's attention, then remains responsible for the ones it misses.

Kindo is financing a broader version of that operating relationship. The Venice, California company announced a $30M expansion financing on November 12, 2026, co-led by Crosspoint Capital and Forgepoint Capital, bringing cumulative capital raised to $57.6M. The money will expand its commercial organization and accelerate development of its enterprise AI Control Center and Deep Hat products.

The investment puts backing behind a shared workplace for IT, Security, Finance and Compliance. As agents gain permission to use company systems, the people approving that access also need a practical relationship with the people supporting the software and paying for its model calls. Kindo is trying to make those responsibilities visible in the same operating record.

What the $30M financing changes for Kindo

The new financing gives Kindo more capacity to sell and develop an enterprise agent management platform. Crosspoint Capital and Forgepoint Capital co-led the investment, with TFX Capital and CircleRock Capital participating. Returning investors include Drive Capital, Riot Ventures, RRE Ventures, Eniac Ventures, Marlinspike Partners, New Era Ventures and Sunset Venture. The company has described the transaction as an expansion; it has not assigned a Series designation or disclosed a valuation.

The financing also brings announced board additions: Andre Fuetsch, a managing director at Crosspoint Capital, and Ernie Bio, a managing director at Forgepoint Capital. Their arrival adds people from the investment firms to the decisions about how widely the product should reach and which enterprise needs deserve development attention. The announced $57.6M is the company's cumulative financing, with $30M representing the new capital.

Founder and CEO Ron Williams brings prior CIO and CISO experience at Riot Games, Clover Health and Bird. Kindo's current leadership page identifies Charlie Hulcher as co-founder and Head of Technology, and Tony Wong as co-founder and SVP of Services; its company facts also name Bryan Vann and Margaret Jennings as co-founders. That combination of technology and service leadership matters commercially. A buyer letting agents into established systems needs the software connected, the policies understood and somebody available when a run needs intervention.

Where Kindo can control an AI agent

Kindo's platform documentation draws a useful boundary between governing requests and controlling the execution of an agent. External agents can route connected model and data calls through Kindo's gateways. The platform associates those requests with identities, applies access policies and records the resulting decisions and costs. That gives an enterprise a point at which to inspect and govern the calls passing through it, while keeping the agent builder it already uses.

Agents built or ported into Kindo's own harness have a deeper operating relationship with the platform. The company describes approval gates, pause and resume controls, and human intervention during execution. A record of the model request and a way to interrupt the running workflow answer different operating questions. Buyers need to understand which controls apply to the agents they intend to connect and which require bringing the work inside the harness.

The deployment choices are similarly consequential for procurement. Kindo describes operation in its cloud, a customer's cloud, on premises or in an air-gapped environment, giving teams options about where the system runs. Those options still leave implementation work: connect the right systems, establish owners and decide which actions require approval. For an enterprise security team, that work is part of buying the product. The control system has to reflect the institution's authority closely enough for people to depend on its decisions when an agent starts touching operational data.

Why the AI budget reaches the security desk

Kindo's AI cost-management description starts with an accounting problem that shared credentials can obscure: who spent the money, through which agent, on what work. The company describes attribution by agent, owner and team, with budget alerts and policies governing model access. An expensive model choice becomes something an organization can discuss alongside the identity and permissions of the person using it. A spending record explains the cost side of enterprise AI return measurement; judging the return also requires understanding the outcome.

Putting those records together can change the purchasing conversation. Finance needs a defensible explanation of spending; security needs a defensible explanation of access. The same unattended workflow can create both questions. A model call that retries repeatedly consumes money while the task remains unfinished, and an agent with broad access can keep working beyond the point where a human expected to review it. These are operating scenarios the described controls address, rather than measured outcomes across Kindo's customers.

Kindo reports approximately 800% year-over-year revenue growth in 2026, without disclosing the revenue base. That company-reported figure indicates commercial momentum within its own account, while leaving the size of the business undisclosed. The go-to-market investment will put more resources behind explaining the product to several enterprise constituencies at once. A platform joining their records also has to accommodate their different reasons for caring about the same agent.

The human decision inside cheaper agent operations

Deep Hat Sift makes the cost discussion concrete. Kindo describes a model that learns from an agent's verdicts which events the agent would dismiss. It begins in shadow mode, observing rather than withholding, so the team can inspect its behavior before a person approves the filtering threshold. After approval, withheld events are logged and a sample continues through for review of missed cases.

That creates a different operating decision from choosing a cheaper model for every event. The team is deciding when an agent should run at all. The dedicated product page avoids a universal accuracy or savings figure, appropriately leaving performance tied to the customer's events and the threshold it accepts. A lower bill and an acceptable miss rate have to be considered together by the people responsible for the queue.

Williams's account of the financing puts commercial expansion alongside deeper control and cost-management development. That is a demanding combination to take into the next customer meeting: the product must give several teams enough shared information to make decisions while respecting the systems they already operate. As Kindo broadens that work, each approved threshold leaves a person with a continuing assignment to watch the sample, revisit the policy and decide when the agent should receive more of the queue again.