# Mendral

Aiolum story: https://aiolum.com/story/mendral/ · added 2026-11-05 · updated 2026-11-06

**In short.** Mendral (AI DevOps agent for CI failures and flaky tests, United States): Team joined Anthropic, July 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders; Y Combinator lists a team of 2 (read November 2026). Launched 26 January 2026; closed later.

AI DevOps agent for CI failures and flaky tests · United States · AI-native · team of 2 to 10 · closed · group: Developer tools and coding agents

- Outcome: Team joined Anthropic, July 2026 (evidence grade of the headline number: E; no revenue figure)
- Launch: 26 January 2026; months from launch to the outcome: 6
- Team: 2 co-founders; Y Combinator lists a team of 2 (read November 2026)
- Outside funding: Y Combinator, Winter 2026 batch, and other investors; amounts not published
- Main factor: Founders' CI expertise
- Meters, 1 to 4: evidence strength 1, copyable pattern 1, capital needed 2
- Tags: AI-native, Small team, Funded, Closed, B2B, Developer tools

Links:

- Visit mendral.com: https://mendral.com/

**The product.** An AI agent for software teams that worked on everything around the code after a pull request was opened. It was installed as a GitHub App, read the logs of GitHub Actions builds, found why a build failed, quarantined or fixed flaky tests, made slow builds faster, reviewed dependency updates for security problems and opened its own pull requests with fixes. It could also post answers in Slack.

**Why it worked.** The two founders had built and scaled CI systems at Docker and Dagger for more than ten years. They wrote that coding agents make teams produce more code, so more builds run and fail, and that the people who write the code with AI tools are not the ones who debug the failed builds. Mendral was the specialist agent for that work. In July 2026 the team joined Anthropic and the hosted product was wound down.

Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; E = no revenue figure found, so the headline is a non-money fact and the other facts carry their own grades; calc = computed on this page from graded facts.

## Summary

Mendral was an AI DevOps agent. DevOps is the work of building, testing and releasing software, and CI (continuous integration) is the automatic build and test run on every code change. A flaky test is a test that sometimes passes and sometimes fails on the same code. Mendral read the CI logs, found the cause of each failure and opened a pull request with a fix, with no person in the loop.

Sam Alba and Andrea Luzzardi, who were early engineers at Docker and then co-founded Dagger, founded Mendral in 2025 and posted its Y Combinator launch on 26 January 2026. That post said Mendral managed CI/CD in production for 15 teams, 5 of them paying customers, including the analytics company PostHog. The product ran on Anthropic's Claude models.

On 21 July 2026 the founders wrote that "The Mendral team is joining Anthropic" and that they were winding down the hosted product. They named no price and no deal terms, and no revenue amount was found in any source, so this story has grade E: a real business with checked facts but no revenue figure.

## Pricing

Mendral's pricing page, read 5 November 2026 (grade C); it published no amounts

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| One plan, flat monthly rate |  | quote | priced by number of engineers; all repos and CI pipelines, no usage caps |

The page said: "One plan, flat monthly rate. No per-seat surprises." A visitor chose a team size (1-10, 11-50, 51-200, 200+) and asked for a quote. No price appeared on any page we read, so both price columns are empty.

## How much revenue did Mendral make before it closed? The numbers, graded

A two-founder San Francisco company whose team joined Anthropic six months after its public launch; no revenue amount or deal terms are public

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Status, 21 July 2026 | team joined Anthropic; hosted product wound down | C | The founders' post: "The Mendral team is joining Anthropic" and "We're winding down the hosted Mendral product and working directly with each customer to ensure a smooth transition." No price or deal terms are given. | [2] |
| Y Combinator status, 5 November 2026 | Inactive | C | Y Combinator company page data: "ycdc_status":"Inactive". | [3] |
| Customers, 26 January 2026 | 15 teams in production, 5 of them paying | C | Launch post: "Mendral is managing CI/CD in production for 15 teams. 5 of them are paying customers, including PostHog". | [4] |
| Revenue amount | not published | C | No source gives a revenue amount, a price or a sale price. | [2, 4, 9] |
| Founders | Sam Alba (CEO) and Andrea Luzzardi (CTO) | C | About page: Sam Alba was VP of Engineering at Docker and a co-founder of Dagger; Andrea Luzzardi was an early Docker engineer and a Dagger co-founder. | [5] |
| Team size and founding year | 2, founded 2025 | C | Y Combinator company page: team size 2, year founded 2025, San Francisco. | [3] |
| Funding | Y Combinator, Winter 2026 batch | C | About page: "Backed by Y Combinator (W26)". The closing post also thanks "our investors". No amount was found. | [5, 2] |
| Usage claims | 16,000+ CI investigations a month; 134 seconds median time to diagnose | C | Company statements on the home page. The February 2026 post says PostHog alone ran 575K CI jobs a week through Mendral. | [1, 6] |
| Named customers | PostHog, Metabase, Luminai | C | Home page: "now deployed at PostHog, Metabase, Luminai, and several other teams." | [1] |
| Model provider | Anthropic (Claude Haiku, Sonnet, Opus) | C | February 2026 post: "We use all three Anthropic tiers (Haiku, Sonnet, Opus)." | [6] |
| Time from launch post to the team's move | 6 months | calc | Calculated: January 2026 to July 2026. | [2, 4] |

How to read this table: a grade C value is a statement by the company on its own pages or its Y Combinator posts. A "calc" value was computed on this page. No value is grade A or B, and no revenue amount exists in the sources.

## When did Mendral launch, and how did it grow? Timeline

From a Y Combinator launch to the team joining Anthropic, in 6 months. Each entry carries its grade.

- **2025** (grade C [3, 5]): Sam Alba and Andrea Luzzardi found Mendral in San Francisco.
- **26 January 2026** (grade C [4]): Launch post in Y Combinator's Winter 2026 batch: 15 teams in production, 5 paying customers, including PostHog.
- **10 February 2026** (grade C [7]): Blog post on PostHog's CI: 575,894 jobs and 1.18 billion log lines in one week.
- **26 February 2026** (grade C [6]): Blog post on how the agent is built: a Go backend, ClickHouse for logs and all three Claude model tiers.
- **21 July 2026** (grade C [2]): The founders announce that the team is joining Anthropic and that the hosted product is being wound down.
- **5 November 2026** (grade C [1, 3]): Y Combinator lists the company as Inactive; the site still shows the product with a banner about the move.

## Why did Mendral work?

### The data, not only the model

The founders wrote that a general coding agent cannot see why a build fails, because the signal is in this run's logs, the last 50 runs and the test history. Mendral loaded billions of CI log lines a week into ClickHouse, a database for fast queries over large tables, and its agent wrote its own SQL queries to investigate.

### Cheap model first, expensive model when needed

A small model (Claude Haiku) read logs and decided whether a failure was a duplicate of a known one; only harder cases went to the largest model (Claude Opus). The March 2026 post says 4 out of 5 failures never reached Opus.

### Work done, not suggested

The agent opened pull requests with fixes instead of sending alerts. The home page pointed to more than 200 merged Mendral pull requests in PostHog's public code repository as evidence.

## How it was built

The founders described the system in their blog. These points are company statements, grade C.

- Anthropic's Claude models in three sizes: Haiku for log parsing, Sonnet for collecting facts, Opus for root-cause analysis and fixes.
- An agent loop written in Go, without LangChain, LangGraph or another agent framework.
- CI logs stored in ClickHouse, compressed at 35:1, as the February 2026 post says.
- Firecracker microVMs on Blaxel as sandboxes for running tests and checking fixes; Inngest for durable execution, so a failed step is retried without repeating the whole investigation.
- Installed as a GitHub App for GitHub Actions; Slack for questions and answers.

Sources: [6, 8, 5]

## Who tried this before Mendral, and why did they fail?

Three companies in our database of failed startups sold tools for building, checking or running developers' code and failed or were absorbed. The values below are from that database and are not independently verified.

The failed startups come from our database of dead startups, retold here in our own words; Its entries are AI-assisted summaries of public sources, so they carry no evidence grade.

### Sider (2012 to 2024; funding US$7M; cause: Competition)

- Promise: Automated code review and static analysis for development teams, run in their CI pipelines.
- Why it died: It was squeezed between free open-source tools and large platforms that bundled code analysis into broader suites.
- Record: 

### AppHarbor (2011 to 2014; funding US$500K; cause: Competition)

- Promise: Deploy .NET applications from GitHub without DevOps work.
- Why it died: AWS and Azure offered more complete services, and its bet on one technology stack limited it.
- Record: 

### Fig DevTool (2020 to 2024; funding US$2.4M; cause: Unit Economics)

- Promise: IDE-style autocomplete and visual command builders for the developer's terminal.
- Why it died: Developers liked it but few paid; AWS acquired it in 2024.
- Record: 

## What changed

Sider checked code against fixed rules and produced warnings that a person still had to read and act on. Mendral's bet was that a language model, given the full log history, can do the person's part: find the cause and write the fix.

The same models made the business hard to keep separate. The founders wrote that every few months a new model made part of their roadmap unnecessary, and the model maker itself, Anthropic, took the team. Like Fig, which AWS took in, the product ended inside a larger company rather than as a business of its own.

## Can one person copy Mendral now?

**Verdict: The method, yes. The result, no.**

The method is public and well documented in the founders' blog. But the result rested on ten years of CI work at Docker and Dagger and on access to large customers such as PostHog. One person without that background and those contacts is unlikely to repeat it.

1. Pick one kind of tedious engineering work that you have done for years and know the patterns of.
2. Collect the data the work needs (here, the full CI log history), not only the code.
3. Route each step to the cheapest model that can do it, and keep the largest model for the hard cases.
4. Show the work in public, for example merged pull requests in a customer's open repository.

**What is still open.** The same approach for other operations work around code: database migrations, cloud cost reports, incident follow-ups.

## What could go wrong for Mendral?

- **What is not verified.** No revenue figure, no price and no deal terms were found. We looked for a payment-processor profile, a founder interview with a dated figure, a company filing and an announcement by Anthropic. The only traction figure is the founders' own count of 15 teams in production, 5 of them paying, in their January 2026 launch post. Their July 2026 post says the team is joining Anthropic; it does not say whether Anthropic bought the company or only hired the team, so this story uses the outcome "closed", not "exit". A reader could look for an Anthropic statement or a later interview with the founders.
- **Why it closed.** On 21 July 2026 the founders wrote: "The Mendral team is joining Anthropic" to work on Claude Platform, and "We're winding down the hosted Mendral product and working directly with each customer to ensure a smooth transition." They gave as the reason that there was no better place to work on what software engineering is becoming. Y Combinator listed the company as Inactive on 5 November 2026.
- **Competition.** The home page itself said that Ramp, Cloudflare and Coinbase built such agents internally, and the February 2026 post answers customers who asked why they should not simply point Claude Code at their failing builds. No other story on this site sells an agent for CI work.
- **Dependence on model providers.** The founders wrote that "Claude has been under the hood since our first commit" and that the agent used all three Anthropic model tiers. They also wrote that every few months a new model made part of their roadmap unnecessary. The product depended on one model provider, which in the end hired the team.
- **Customer claims are the company's own.** The usage figures (16,000+ investigations a month, 134 seconds median time to diagnose, 1.18 billion log lines a week at PostHog) and the named customers are stated by Mendral. We did not check the merged pull requests in PostHog's repository.

## Sources

1. Mendral home page: Platform Engineering on Autopilot — https://mendral.com/ (grade C). Read in full on 5 November 2026. Customers, usage claims and the banner about joining Anthropic.
2. Mendral blog, 21 July 2026: The Mendral team is joining Anthropic — https://mendral.com/blog/mendral-team-joins-anthropic (grade C). Read in full. The founders' post on the team's move and the wind-down of the hosted product.
3. Y Combinator company page: Mendral, AI DevOps Engineer — https://www.ycombinator.com/companies/mendral (grade C). Read the page data (downloaded by the pipeline on 5 November 2026): status, team size, year founded, location and batch.
4. Launch YC, 26 January 2026: Mendral - AI DevOps Engineer — https://www.ycombinator.com/launches/PHN-mendral-ai-devops-engineer (grade C). Read in full. Source for the 15 teams and 5 paying customers.
5. About Mendral — https://mendral.com/about (grade C). Read in full on 5 November 2026. Founders, founding year, backers and product description.
6. Mendral blog, 26 February 2026: We use Claude Code daily. We still built our own CI agent. — https://mendral.com/blog/how-we-built-our-ai-agent (grade C). Read in full. Source for the models, the Go backend, ClickHouse, Blaxel and Inngest.
7. Mendral blog, 10 February 2026: What CI Actually Looks Like at a 100-Person Team — https://mendral.com/blog/ci-at-scale (grade C). Read the section on PostHog's CI in one week.
8. Mendral blog, 6 March 2026: We Upgraded to a Frontier Model and Our Costs Went Down — https://mendral.com/blog/frontier-model-lower-costs (grade C). Read the sections on model routing and cost.
9. Mendral pricing page — https://mendral.com/pricing (grade C). Read in full on 5 November 2026. Source for the pricing table; no amounts are published.
10. Database of failed startups, local export — /method/#dead (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.

Contact: aiolum@livyden.com. Run Mendral? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: Mendral).