Asymptote Labs, a New York startup that monitors what AI agents do inside companies, has raised $5.5 million in seed funding. Tower Research Ventures, the venture arm of trading firm Tower Research Capital, led the round. Lytical Ventures, 1984 Ventures and South Park Commons also took part. The company did not disclose its valuation.
The funding lands at a moment when enterprises are handing real work to autonomous software agents, and then discovering that they have very little visibility into what those agents actually did. Asymptote is betting that a new layer of monitoring and control software will become as standard inside companies as endpoint security or cloud logging.
Beacon: a flight recorder for AI agents
Asymptote's main product is Beacon, a free, open-source tool that records what AI agents do. It works with coding agents such as Claude Code, Codex and Cursor, plus more than 20 others in the broader agent ecosystem. Beacon captures each agent's activity, including the prompts it receives, the commands it runs and the files it edits, and puts all of it into one standard format.
That standardization is the point. Companies deploying agents today typically run several at once from different vendors, each producing its own logs in its own schema, with its own retention rules. Security teams end up stitching together incompatible records, or simply giving up on the audit trail entirely. Beacon gives them one place to see what different agents did and, critically, on whose behalf they did it.
The open-source distribution model is deliberate. Agents are proliferating faster than procurement cycles can keep up, which means a tool that requires a contract, a security review and an integration project will lag behind adoption. A free tool that a developer can install in minutes spreads through the same bottom-up motion that agent frameworks themselves have used.
The paid layer: blocking actions before they happen
Asymptote's paid platform builds on Beacon and goes beyond recording. The company says it can block risky actions before they occur, citing data leaks as a primary example. In practice, that means shifting from forensic logging, which tells you what went wrong after the fact, to enforcement, which stops the agent mid-task when it tries to reach a resource it should not touch.
The distinction matters commercially. Observability tools are useful but easy to defer; policy enforcement tools sit directly in the path of a task and carry a clearer risk story for a buyer. The paid platform effectively turns Beacon into the sensor layer for a control plane, with the open-source tool generating the telemetry that the commercial product acts on.
Founders with a security background
Asymptote was founded by Justin D'Souza, who serves as chief executive, and Shukan Shah. Both previously worked as engineers at Doppel, a startup that fights social engineering attacks. That background is relevant: the core problem with autonomous agents is not unlike the core problem with phishing, in that the risk arrives through a legitimate-looking channel and the defender has to make a judgment call at machine speed.
D'Souza has framed the opportunity in structural terms, arguing that AI agents are becoming a new operating layer inside every company while the infrastructure to understand and control them has not caught up. The pitch is that agents need the equivalent of what identity, logging and access control became for human users and for cloud workloads.
Traction and early users
Beacon launched in May, and the company says its GitHub page grew its stars by more than 460 percent in September, a signal of developer interest that open-source infrastructure projects often use as an early proxy for adoption.
Asymptote says security teams already use its tools in production, though it did not name customers. Tower's own security team is testing Beacon on its computers, the company said, giving the lead investor a direct look at how the tool behaves on a real corporate fleet rather than a demo environment. That kind of internal pilot is a common step for venture arms investing in infrastructure that their own organization would need to buy.
A crowded and fast-moving category
Asymptote is not alone in chasing the agent governance market. Straiker raised $64 million in June to test and monitor AI agents, a far larger round that signals investor appetite for the category. Larger observability and security vendors are also extending existing products toward agent telemetry, and the major model providers ship their own logging and permissioning features.
The competitive risk for a seed-stage company is that agent monitoring becomes a feature bundled into platforms that companies already buy. Asymptote's counter is partly architectural: by supporting more than 20 agent tools and standardizing their output, Beacon aims to be neutral ground across vendors, in the same way that open telemetry standards became more valuable than any single vendor's proprietary format.
The threat landscape is also moving quickly. This month, researchers said a single prompt let them take over agents on Amazon's AgentCore, though AWS disputed the findings. Regardless of how that specific dispute resolves, the episode illustrates the underlying problem: agents that can act on the world are exposed to manipulation through the very inputs they are designed to trust, and the blast radius of a successful attack depends on what the agent was permitted to do.
Why the timing works
Agent adoption inside enterprises has shifted over the past year from pilots to production work, particularly in software engineering, where coding agents now open pull requests, refactor repositories and run tests. That shift creates two simultaneous needs. Compliance and security teams need a defensible record of what happened, which is hard to produce when agent activity lives in a vendor's dashboard. Platform teams need a way to constrain what an agent can touch, so that a mistake does not become an incident.
Both needs intensify as agents are given credentials, repository access and the ability to call internal services. The principle of least privilege, long applied to human employees and to service accounts, has no clean equivalent in a world where an agent may legitimately need broad access to complete a task and may be manipulated into misusing it. Recording, attributing and then restricting agent behavior is the sequence Asymptote is selling.
What the money is for
Asymptote said the $5.5 million will go toward engineering, hiring and sales. For a company with an open-source core, that allocation is telling: engineering to keep pace with a rapidly changing set of agent tools, hiring to build out both the open-source community effort and the enterprise sales motion, and sales to convert the developer mindshare that Beacon generates into paid platform contracts.
The round also gives Asymptote runway to answer the question that will define the category over the next few years, which is whether agent governance settles into a single standardized layer that every company adopts, or fragments into a set of vendor-specific controls attached to whichever agent platforms dominate. Tower Research Ventures is wagering that a neutral, open-format recorder with an enforcement layer on top is a durable position to hold as that question resolves.