Problem

Modern AI agents do more than generate text: they send emails, modify code repositories, and interact with external APIs, so an incorrect action can cause real operational damage. The core pain is that conventional testing often cannot reproduce the stateful, real-world environment in which an agent failed. The economic cost is the combination of production risk, regressions, and engineering time spent diagnosing and recreating failures. A representative use case is an AI support agent that gives an incorrect answer about a refund window or mishandles a tool call.

Product / Service

Archal is an evaluation and testing platform for AI agents and autonomous software. It captures and grades agent traces, recreates failures inside controlled environments or stateful clones of services such as GitHub, Slack, and Stripe, and then uses a separate coding agent to provision a fix pull request. The fixed agent is rerun against the failed environment, and the failure is stored as a regression test so it does not recur.

The delivery model combines scenario-as-code, sandboxed service clones, trace capture and replay, and CI integration. Teams write scenarios in Markdown, run them against cloned APIs with persistent per-run state, and can fail builds when regressions appear. Archal offers a free tier, a $199-per-seat monthly Pro/Teams plan, and custom Enterprise plans, with usage metered primarily through session-minutes and evaluations.

Market

Archal competes in the AI-agent evaluation and testing market, overlapping with developer tools, API infrastructure, and observability platforms. Its closest named alternatives include LangSmith and Vercel AI Playground, while LangSmith and Braintrust represent adjacent agent-evaluation platforms. Archal’s differentiation is its focus on verifying real actions in stateful external-service environments and closing the loop by generating fixes, rather than stopping at tracing, scoring, or monitoring.

As of August 1, 2026, Archal appears to be an early-access, pre-scale company rather than one with publicly demonstrated customer or revenue traction. Y Combinator lists it as an active Summer 2026 company founded by Noah Song and Aidan Tiruvan with a two-person team in San Francisco, and the company says it is backed by Y Combinator. Its public pricing indicates an intention to commercialize, but the available research does not disclose customer counts, usage, ARR, or revenue, so it is more accurate to describe Archal as early commercial/pre-revenue status not yet publicly established than to claim proven product-market fit.