Asymptote Labs Raises $5.5M Seed to Monitor and Control Enterprise AI Agents

AI agents are moving beyond generating suggestions to editing files, running commands, and working through connected business tools. For security teams, the resulting question is increasingly operational: what did an agent actually do, under whose authority, and can the next risky action be stopped? Asymptote Labs is building infrastructure around that problem and has announced an oversubscribed $5.5 million seed round.

The financing is led by Tower Research Ventures, the venture arm of Tower Research Capital, with participation from Lytical Ventures, 1984 Ventures, and South Park Commons. Asymptote says the capital will support engineering, product development, hiring, and go-to-market expansion.

Its approach combines Beacon, an open-source layer for recording agent activity, with an enterprise platform for detection, response, and governance. The distinction is important: collecting a record of an action and enforcing a decision before that action proceeds are separate capabilities. A credible security architecture needs to make clear which component does each job.

A Common Record Across Different Agent Tools

Beacon captures activity across Claude Code, Codex, Cursor, and more than 20 other agent harnesses, according to the company. A harness is the software environment that connects a model to tools, instructions, and a working session. Different harnesses expose different information, making it difficult to investigate activity consistently across a company’s chosen tools.

Asymptote describes Beacon as an MIT-licensed telemetry layer that normalizes those records into a common runtime data model extending OpenTelemetry. That gives security and engineering teams a shared structure for examining activity instead of interpreting every tool’s history separately.

The Beacon repository adds useful implementation context. It describes session replay covering prompts, responses, tool use, commands, edits, approvals, and token information, alongside local-first JSONL records and explicit destinations. It also describes carrying reviewed knowledge forward through MCP or Agent Skills.

The practical value of normalization is correlation. An investigation might need to connect a user request to a command, a file change, and an approval. A final response alone cannot explain that sequence. A shared event structure can make the sequence easier to inspect, although the completeness of the record still depends on what a particular integration captures.

Where Enterprise Control Enters the Picture

Asymptote’s enterprise platform builds on Beacon with runtime detection and response. The company says it can intercept live agent activity and apply policy-driven guardrails at prompt and tool levels, identifying risks such as data leakage, insider threats, and AI supply-chain issues.

The company website makes the product split more explicit. Open-source Beacon provides portable telemetry that customers can route into their existing security systems. Managed Asymptote adds ingestion, retention, search, detections, identity mapping, policy gates, approvals, and governance workflows. Private deployments are also part of the stated offering.

That separation matters when evaluating a rollout. Recording a tool call after execution helps with investigation; preventing it requires a control point in the execution path. Buyers need to establish which actions can be intercepted, which are observed afterward, and how enforcement behaves when a service is unavailable. The funding announcement does not publish a coverage matrix or a benchmark for blocking accuracy.

These are central questions in enterprise AI agent governance. A policy written in a model’s instructions is different from a policy enforced by infrastructure outside the model. The useful test is whether an agent can proceed with an unauthorized action despite the intended restriction.

Telemetry Can Also Preserve Useful Engineering Knowledge

Beacon’s positioning extends beyond security. Its public project description explains how session history can be used to identify useful workflows, corrections, debugging patterns, and repository conventions for reuse in later sessions. Instead of discarding a solved problem when a conversation ends, the aim is to make reviewed knowledge available to a future agent.

That creates a second reason to collect activity consistently across tools. A workflow discovered in one harness may be valuable to a team using another. However, preserving a successful sequence does not mean it is safe to repeat unchanged. Credentials, dependencies, permissions, and business requirements can change between sessions.

The same records that help diagnose failures may contain sensitive prompts, source code, or tool output. Deployment therefore raises familiar information-management questions: where records go, how long they remain, who can retrieve them, and how sensitive content is handled. Ownership of telemetry is valuable, but it does not eliminate the need to govern access to it.

Founders With a Background in Detection Infrastructure

Asymptote was co-founded by Justin D’Souza, CEO, and Shukan Shah, CTO. Both previously worked at social-engineering defense company Doppel. The company’s About page describes D’Souza’s work scaling machine-learning infrastructure and Shah’s work on security-operations automation and agentic threat hunting.

The announcement says Beacon’s GitHub stars grew more than 460% in September and that the project reached No. 3 on GitHub’s global trending repositories. Those are company-reported adoption signals, rather than evidence of detection effectiveness or recurring revenue. Asymptote also says security teams are using Beacon and its enterprise platform in production, without disclosing customer counts or contract values.

Tower Research Ventures says its interest began with Beacon’s open-source traction and that Tower’s security organization is evaluating the technology on its own endpoints. Evaluation by an investor’s organization should be distinguished from a disclosed commercial customer deployment.

The Next Test Is Coverage and Useful Intervention

The seed round supports a broader thesis: as organizations deploy more agents, runtime records become necessary for security, engineering reliability, platform oversight, and cost analysis. A common data layer could let those teams examine the same activity for different purposes.

The test for Asymptote will be whether that shared view is complete and timely enough to support action. Security teams will need useful detections with manageable false positives, while engineers will need records that explain failures without creating excessive collection overhead. The financing gives the company resources to pursue those goals; its long-term value will depend on what enterprises can reliably observe, investigate, and control.