What does Archal actually build for AI agent testing?

Archal, a YC S26 startup founded in 2026, builds hosted stateful clones of real SaaS APIs like GitHub, Slack, Stripe and Linear so AI agents can be tested against realistic environments, plus an Autoloop feature that turns production traces into failure cases and auto-opens fix pull requests, per its docs dated 2026-06-26.

Is Archal profitable or generating revenue in 2026?

No public revenue, ARR, or customer-count figures exist for Archal as of 2026-07-20. Third-party tracker Extruct.ai lists it only as a 1-10 employee eval platform with no disclosed traction metrics, consistent with a pre-Demo Day YC S26 company.

Could OpenAI replace what Archal does?

OpenAI published its own Agent Improvement Loop cookbook on 2026-06-29, covering traces, evals, harness optimization and Codex handoff, functionally overlapping Archal's core trace-to-fix-PR Autoloop, and OpenAI acquired eval tool Promptfoo, announced 2026-03-09.

Who is Archal's biggest competitor?

Braintrust is Archal's closest well-funded rival, having raised a Series B in February 2026 reportedly reaching an $800M valuation with customers like Notion and Replit, per a June 2026 vendor comparison. Archal, by contrast, has just 2 employees and no disclosed funding.

Archal is privately held, classified by Pitchbook as accelerator/incubator backed with no priced seed or Series A round found as of 2026-07-20. It went through Y Combinator's Summer 2026 batch and is cofounded by Aidan Tiruvan and Noah Song.

What makes Archal different from Braintrust or LangSmith?

Archal's stated differentiator is a library of 20-plus stateful SaaS API clones (GitHub, Stripe, Slack, Datadog, etc.) that preserve referential integrity and real error shapes, per its docs dated 2026-06-26, unlike static mocks used elsewhere. This integration asset is unproven as a lock-in moat against better-funded rivals.

YC Summer 2026 · United States

Archal

Two-person startup building stateful SaaS clones to test and auto-fix AI agents.

Losing its protection

Assessed 27 July 2026 · Evidence as of 27 July 2026

The structural protection is eroding.

On the recordSince 27 July 2026 · Checkable from July 2026 · Every call on the record Checkpoint: OpenAI's Promptfoo acquisition is set to close

The call

Archal has a real, unusual product but operates in a crowded lane where OpenAI and far richer rivals could add the same clone-based fix loop at any time.

Archal builds software that creates fake versions of services so AI agents can be tested safely before going live. The problem is that Braintrust and Galileo are already doing similar things with much more money, and OpenAI just added its own testing features in mid-2026 and bought Promptfoo to strengthen that offering. Archal's main advantage is its library of ready-made clones, but there is no clear evidence that this alone keeps customers locked in or that larger competitors cannot copy it quickly. The company's fate depends on whether it can prove its clone library becomes so essential that customers stick around even after OpenAI and Braintrust add their own versions.

Why

AI pressure: OpenAI. OpenAI shipped its own Agent Improvement Loop in June 2026 and is buying Promptfoo, aiming at the same fix-the-agent job. Could AI do the core job? Yes, likely.

Bigger players: Braintrust. Braintrust, fresh off an ~$800M Series B, already runs evals for Notion, Replit, Cloudflare, Ramp, and Dropbox in this same space. Could a giant copy it? Yes, easily.

What protects it. Its library of 20+ stateful SaaS clones (GitHub, Slack, Stripe, Linear, Jira, etc.) with real referential integrity is unusual and slow to copy well.

What would change this call

OpenAI finalizes its Promptfoo acquisition and folds it into the Agent Improvement Loop by late 2026. Shows the model layer itself closing the gap Archal's product targets.
Braintrust announces a stateful SaaS clone or sandbox feature in a product update. Shows the incumbent fast-following Archal's core differentiator.
Archal discloses a paying customer, CI-gated usage, or a seed/Series A raise within a year. Shows whether Archal is gaining real traction beyond trial usage.

Key numbers

Braintrust Series B at ~$800M valuation (2026-02)
Galileo has raised $68M total (2026-04-12)
Archal: 2 cofounders, no disclosed revenue (2026-07-20)

Named risks

OpenAI's native Agent Improvement Loop cookbook duplicates Archal's core trace-to-fix-PR workflow. Severity: high
Well-funded incumbents like Braintrust could build a comparable stateful SaaS clone library within months. Severity: high
No public revenue, customers, or funding disclosed as of 2026-07-20, leaving runway and traction unverified. Severity: medium
Only 2 employees with no evidence of hiring growth as of July 2026. Severity: medium
No independent third-party validation exists of Autoloop's defect-catch effectiveness. Severity: medium

Rivals, as read in July 2026

Rival	Direction	Evidence
Braintrust	losing to	Braintrust raised Series B reaching reported ~$800M valuation Feb 2026 vs Archal's 2-person, unfunded stage
OpenAI	losing to	OpenAI shipped native Agent Improvement Loop cookbook (2026-06-29) replicating Archal's trace-to-fix-PR workflow
LangSmith	contesting	LangSmith offers established $39/seat/month pricing and LangChain ecosystem lock-in as of 2026-05-21
Latitude	contesting	Latitude markets an issue-to-PR loop via MCP server, closest positioning overlap with Archal's Autoloop, as of 2026-07-01
Lemma	contesting	YC-sibling Lemma (F25) pitches an eval-to-auto-optimization loop, a close conceptual peer to Archal

Questions people ask

What does Archal actually build for AI agent testing?: Archal, a YC S26 startup founded in 2026, builds hosted stateful clones of real SaaS APIs like GitHub, Slack, Stripe and Linear so AI agents can be tested against realistic environments, plus an Autoloop feature that turns production traces into failure cases and auto-opens fix pull requests, per its docs dated 2026-06-26.
Is Archal profitable or generating revenue in 2026?: No public revenue, ARR, or customer-count figures exist for Archal as of 2026-07-20. Third-party tracker Extruct.ai lists it only as a 1-10 employee eval platform with no disclosed traction metrics, consistent with a pre-Demo Day YC S26 company.
Could OpenAI replace what Archal does?: OpenAI published its own Agent Improvement Loop cookbook on 2026-06-29, covering traces, evals, harness optimization and Codex handoff, functionally overlapping Archal's core trace-to-fix-PR Autoloop, and OpenAI acquired eval tool Promptfoo, announced 2026-03-09.
Who is Archal's biggest competitor?: Braintrust is Archal's closest well-funded rival, having raised a Series B in February 2026 reportedly reaching an $800M valuation with customers like Notion and Replit, per a June 2026 vendor comparison. Archal, by contrast, has just 2 employees and no disclosed funding.
Who owns Archal?: Archal is privately held, classified by Pitchbook as accelerator/incubator backed with no priced seed or Series A round found as of 2026-07-20. It went through Y Combinator's Summer 2026 batch and is cofounded by Aidan Tiruvan and Noah Song.
What makes Archal different from Braintrust or LangSmith?: Archal's stated differentiator is a library of 20-plus stateful SaaS API clones (GitHub, Stripe, Slack, Datadog, etc.) that preserve referential integrity and real error shapes, per its docs dated 2026-06-26, unlike static mocks used elsewhere. This integration asset is unproven as a lock-in moat against better-funded rivals.

Timeline

2026-07-20YC lists Archal team size at 2 cofounders, San Francisco based
2026-07-09Archal ships CLI quickstart docs (npx archal init, npx archal run)
2026-07-06Archal ships clone-session CLI documentation
2026-07-04Archal publishes CI integration guide for GitHub Actions and GitLab CI
2026-06-29OpenAI publishes native Agent Improvement Loop cookbook, overlapping Archal's Autoloop
2026-06-26Archal introduction docs describe stateful SaaS clone library
2026-06-11Comparison source describes Braintrust's Feb 2026 Series B at reported ~$800M valuation
2026-03-09OpenAI announces acquisition of eval tool Promptfoo
2026Archal founded, joins Y Combinator Summer 2026 (S26) batch

Revision history

27 July 2026

First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.

In the same band

Cosmic Robotics Pango Atlas Discovery