About Asymptote Labs
Asymptote Labs sells a runtime data platform for AI agents. Asymptote gives teams one place to understand agent activity, investigate behavior, enforce controls, and analyze usage across tools and environments. Asymptote provides enterprise security and engineering teams a unified AI telemetry layer to see what agents are doing, govern how they operate, and continuously improve them over time.[3]
Asymptote says agents are spreading across tools and teams faster than organizations can track them. Asymptote says agents can access data and take actions without enough real-time oversight or control. Asymptote says teams can't see which agents are effective, failing, or wasting spend.[3]
Beacon is the open-source part of the offering. Asymptote is powered by Beacon, an open-source telemetry layer for capturing and normalizing agent activity across every environment. Agent Beacon is the open-source telemetry foundation, and the Asymptote platform builds on it with the managed control enterprises need to govern agent activity at scale. Agent Beacon captures activity across 20+ agent harnesses including Claude Code, Cursor, Codex, and OpenClaw. Agent Beacon forwards normalized telemetry to Splunk, Datadog, CrowdStrike, and other existing systems. Agent Beacon stores a portable local activity record in JSONL. Beacon gives Security and IT teams local endpoint visibility into supported agent harnesses. Beacon normalizes captured activity into a stable endpoint event schema and writes JSONL for local inspection or customer-managed forwarding.[3][4][5]
Beacon is deployed to endpoints in several ways. Asymptote deploys Beacon across endpoints with Jamf, Fleet, and other MDM workflows. Beacon macOS deployment uses a signed and notarized package distributed through Jamf Pro, Fleet, or another MDM. On Linux, the Beacon collector runs as a systemd service. Beacon browser chat telemetry is limited to the Claude.ai and ChatGPT conversation streams captured by the optional Agent Beacon browser extension.[4][5][6][10]
Asymptote Platform is the managed layer. Asymptote Platform is a managed runtime data platform for operating AI agents across the organization. Asymptote Platform includes managed ingest, retention, search, and analytics. Asymptote Platform includes fleet-wide agent inventory and usage visibility. Asymptote captures prompts, tool calls, file edits, approvals, and runtime events. Asymptote inventories Claude Code, Codex, Cursor, and other agent harnesses. Asymptote normalizes activity into one consistent source of agent runtime data.[3][4]
Asymptote builds a complete inventory of agents and the environments they run in across an organization. Asymptote detects malicious and suspicious agent behavior in real time, with the full execution context needed to understand what happened. Asymptote surfaces threats like secret access, prompt injection, malicious tool use, suspicious commands, and data exfiltration across the complete agent session. Asymptote Guard steers behavior, blocks actions inline, revokes access, contains sessions, or terminates agents when needed. Asymptote adds visibility into the agent runtime, including the harness, prompt context, tool calls, resources accessed, and policy decisions behind each action.[1][7]
Asymptote correlates every action in an agent session so engineers can pinpoint failed commands, retries, tool errors, slow steps, and execution loops. Asymptote lets engineering teams understand which agents, models, and workflows developers rely on across the organization. Asymptote connects model spend to the teams, agents, users, and workflows that generated it. Asymptote adds runtime context, connecting AI spend to specific users, agents, repositories, sessions, and workflows. Asymptote Guard enforces model, budget, and session-level controls so teams can limit expensive usage, stop runaway workflows, and keep AI spend within policy.[8][9]
Asymptote can be self-hosted in AWS, Azure, or GCP for tighter control over infrastructure, networking, and data residency. Asymptote offers dedicated infrastructure for organizations that need tighter control over how agent data is stored and operated. Asymptote integrates with providers like Okta for secure access without additional passwords. Asymptote protects data with AES-256 for storage and backups and TLS 1.2+ for all data in transit. Asymptote enforces zero-day retention with providers such as OpenAI and Anthropic, ensuring customer data is not stored or used for training. Asymptote connects to the tools teams already use and extends with custom actions for workflows and systems unique to a business.[4][10][11]
Asymptote Labs on the record