Companies

Armature

armature.tech

Armature provides analytics, testing, and optimization infrastructure for products used by AI agents.

HQSan Francisco, California, United States
Employees1-50
Jobs checked 17h ago
Developer ToolsAI InfrastructureAPI / Platform

About

Armature builds analytics, automated testing, and optimization infrastructure for companies whose products are used by AI agents. Its platform captures agent sessions occurring inside clients such as Claude, including workflows and reasoning traces that traditional UI analytics cannot see, helping teams improve MCPs, CLIs, and agent-facing products.

Market

Armature competes in the emerging Agent Experience (AX) market: observability, product analytics, testing, evaluation, and optimization for software operated through AI agents. Its positioning is product-side rather than merely developer-side: it captures users’ agent sessions in Claude, ChatGPT, and other clients, identifies intents and failures, replays traces, scores outcomes, and supports benchmarking and remediation. Armature explicitly differentiates from PostHog, Amplitude, and Mixpanel because they focus on humans clicking a UI, and from LangSmith and Langfuse because those tools primarily observe agents built by engineering teams rather than how customers’ agents experience a shipped product.

Target Customers

Armature targets software companies building agent-accessible products through MCP servers, Claude Connectors, ChatGPT Apps, or CLIs, especially product, engineering, and AI-platform teams responsible for agent experience and reliability. The available materials do not identify a narrower industry or employee-size band; the clearest qualification is that customer sessions occur inside external AI clients rather than the company’s own UI.

At a Glance

Problem

As software shifts from human-operated interfaces to AI agents, companies lose visibility into how customers actually experience their products. Sessions increasingly take place inside Claude, ChatGPT, and other agent clients rather than in the company’s own UI, leaving product teams unable to see what users asked, what the agent attempted, where context was missing, or why a workflow failed. The most damaging failures are silent: an agent can loop, reach a dead end, or fail to complete a task even when underlying API responses appear healthy, causing users to abandon the product or churn.

The central use case is making agent-facing products debuggable and improvable: a company needs to discover which real workflows customers are attempting, identify the highest-volume failure modes, and determine whether its MCP server, CLI, connector, or ChatGPT app successfully completes those workflows. This turns an otherwise invisible agent experience into an observable product-quality and retention problem.

Product / Service

Armature is a cloud product-analytics and testing layer for AI-agent sessions. Customers instrument their MCP servers, Claude Connectors, or ChatGPT Apps with Armature’s SDK; the service captures, rebuilds, classifies, and scores the resulting sessions in a dashboard. It groups sessions by user intent and use case, ranks them by volume and success rate, flags failures and loops by root cause, and lets teams replay the full interaction trace to see exactly where a task broke.

The product is expanding beyond analytics into automated testing and benchmarking. Armature runs complex workflows across major agent harnesses and models, evaluates MCP and CLI performance against comparable tasks, and exposes the results through public or private benchmarks. It offers a self-serve free tier with 1,000 sessions per month, then charges $50 per additional 1,000 sessions, with custom plans offering enterprise features such as SSO, audit logs, custom retention, and onboarding.

Market

Armature competes in the emerging Agent Experience market: the intersection of product analytics, AI-agent observability, MCP or CLI testing, and agent-quality benchmarking. Its closest named adjacent competitors are LangSmith and Langfuse, which focus primarily on observing and evaluating the agents that engineering teams build. Armature differentiates by showing product teams how external users’ agents experience the software they ship, including workflows that the company may not yet support.

The company appears to be in an early commercial launch phase rather than having publicly demonstrated mature revenue traction. It is backed by Y Combinator, has publicly launched its product, offers free audits and self-serve usage, and reports 915 agent runs on its benchmark site. The public materials reviewed do not disclose customer counts, revenue, or paid conversion, so Armature should not be described as definitively pre-revenue; its visible traction is product availability, benchmark activity, and early ecosystem adoption rather than disclosed financial performance.

Founders & Leadership

Theodore OtzenbergerFounder
Co-Founder
Louis ScreminFounder
Co-Founder & Founding Engineer

Funding History

2026-05
Pre-Seed$500K

Y Combinator

Recent News

2026-07-21product
Agent reviews — what agents say about the tools they use

Armature introduced an agent-review feature in which AI agents automatically review tools after using them on real tasks through a public review intake.

2026-06-08
Armature listed in The YC Tier List for the Spring 2026 batch

Armature received coverage in The YC Tier List, highlighting co-founder Theodore Otzenberger’s Palantir background and the company’s enterprise AI-agent tooling focus.

2026-05-19product
Armature - Making your app finally usable by AI agents

Armature publicly launched its platform for monitoring, analyzing, and optimizing how AI agents experience products. The launch described automated testing, agent-session tracing, benchmarking, and auto-remediation for MCPs and CLIs, with a free self-serve option.

2026-05-06
Armature joins Y Combinator’s Spring 2026 company cohort

Y Combinator profiled Armature as a Spring 2026 company founded by Theodore Otzenberger and Louis Scremin. The company’s initial product provides analytics for AI-agent sessions and automated testing for MCPs and CLIs.

Active Roles

0

No active roles right now.

Get notified when they post

Business Model

Armature uses a freemium, usage-based SaaS model: its free plan includes 1,000 sessions per month, after which usage costs $50 per 1,000 sessions. It also offers custom plans for larger customers.

Products

Product analytics for AI-agent sessions, including intent/use-case grouping, success scoring, issue detection, and session replayAutomated testing infrastructure for MCPs and CLIs across agent harnesses and modelsAgent-experience analytics and benchmarking against competitorsAuto-remediation through an Armature MCP for internal coding agents

Tech Stack

MCP servers and MCP/CLI workflowsArmature SDK integrations for MCP servers, Claude Connectors, and ChatGPT App backendsTypeScript, Python, and Go instrumentationAI-agent/LLM client integrations including Claude, ChatGPT, Cursor, Codex, and Gemini CLISession tracing, replay, classification, scoring, benchmarking, and detection models

Competitors

LangSmith
Langfuse
PostHog
Amplitude
Mixpanel