Firetiger: Revenue, sale and how it was built
Firetiger (AI agents that check each code change in production and investigate incidents, Not stated in the sources): Acquired by Cursor, August 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders, Rustam Lalkaka and Achille Roussel, from 2024; headcount at the sale not published; the whole team joined Cursor in August 2026. Launched 18 February 2026; sold 6 months after launch.Added 7 November 2026
- Outcome
- Acquired by Cursor, August 2026
- Evidence
- E no revenue figure was checked
- Launch
- 18 February 2026
- Status
- Sold, 6 months after launch
- Team
- 2 co-founders, Rustam Lalkaka and Achille Roussel, from 2024; headcount at the sale not published; the whole team joined Cursor in August 2026
- Outside funding
- US$7.6M seed round led by Sequoia Capital, with angel investors, announced 18 February 2026, the company said
- Main channel
- Not stated
What is not verified
No sale price, revenue figure, customer count or headcount was found. We read Cursor's announcement, Dealroom's news item, and Internet Archive copies of Firetiger's home, pricing and blog pages. Firetiger's own post on the sale (blog.firetiger.com/firetiger-is-joining-cursor/) could not be opened: the blog host's security certificate no longer matches its name, and the Internet Archive had no copy. The home page showed eight customer logos under "Trusted by"; blog case studies support three of them (Kernel, Town, Voxel), and we found no source for the other five (Clerk, Avora, Architect, Scale to Win, Judgment Labs). The launch post's 32% fewer change-related issues and Dealroom's "tens of thousands" of changes come with no method or count. A reader could look for a statement by the lead investor, Sequoia Capital, or a later interview with the founders. This story is grade E: the business is real and its product, price and customers are described from public pages, but no revenue figure was checked. Aiolum shows it as an idea worth studying, not as proof that it makes money.
Firetiger sold AI agents, programs in which a language model works through a task step by step, calls tools and chooses its next step itself, for production, the live systems that customers use. Its main product, Change Monitors, read the diff of a pull request (the lines that a code change adds or removes), wrote a plan of what should change, and after the deploy, the moment the change goes live, compared logs, metrics and traces in each environment and reported whether the change did what it was meant to do. Other agents triaged production problems and passed a suspected root cause to coding agents such as Cursor and Claude Code.
Rustam Lalkaka and Achille Roussel founded the company in 2024, according to Cursor's announcement. It launched publicly on 18 February 2026 with self-serve paid plans from US$599 a month and a US$7.6 million seed round led by Sequoia Capital, the launch post said. On 13 August 2026 Cursor announced that the Firetiger team was joining it, and no price was disclosed. Dealroom reported the next day that the standalone product was being shut down, new signups were closed and all user data would be deleted after each team's end date.
No sale price, revenue figure, customer count or headcount was found, so this story is grade E: the evidence is the buyer's own post, a news item and Internet Archive copies of Firetiger's pages. The product no longer exists: on 7 November 2026 www.firetiger.com answered HTTP 402 with the message DEPLOYMENT_DISABLED.
How much did Firetiger make, and what was it sold for?
Each row shows the evidence grade and the source numbers from the list at the end of this page. The grades are explained in How Aiolum verifies.
| Fact | Value | Grade | How we know | Sources |
|---|---|---|---|---|
| Outcome, 13 August 2026 | acquired by Cursor; the team joined; price not disclosed | A | Cursor's own announcement: "I'm excited to announce that the Firetiger team is joining Cursor." Dealroom, 14 August 2026: "Cursor has acquired Firetiger" and "Terms were not disclosed." | [1, 2] |
| What happened to the product | standalone product shut down; new signups closed; user data to be deleted | C | Dealroom: "the standalone product is being sunset" and "New signups are closed, active teams have been given service end dates, and all user data will be deleted afterward." Firetiger's own post on the sale could not be opened. On 7 November 2026 www.firetiger.com answered HTTP 402 "DEPLOYMENT_DISABLED". | [2, 10] |
| Founded | 2024, by Rustam Lalkaka and Achille Roussel | A | Cursor: "The company was founded in 2024 by Rustam Lalkaka and Achille Roussel." Dealroom says the same. | [1, 2] |
| Team | 2 co-founders and a team; headcount not published | A | Cursor: "We're looking forward to building this with Rustam, Achille, and the rest of the Firetiger team." No source gives the number of people. The blog carries posts by at least two other staff members. | [1, 9] |
| Public launch | 18 February 2026, with self-serve sign-up and free trials | C | Launch post: "Today, we're excited to announce Firetiger." It mentions early access customers before that date. | [3] |
| Funding | US$7.6 million seed round led by Sequoia Capital | C | Launch post, with angel investors including Matthew Prince and Calvin French-Owen. The home page of June 2026 still showed "Announcing Firetiger and $7.6m seed round". | [3, 4] |
| Prices | plans from US$599 a month; from 9 June 2026 US$0.50 per investigation and US$5 per pull request watched | C | Launch post and the archived plan page of 8 June 2026; the pricing post of 9 June 2026 for the usage prices. | [3, 5, 6] |
| Named customers | Kernel, Town and Voxel in case studies; 8 logos on the home page | C | The home page showed the logos of Clerk, Kernel, Avora, Architect, Scale to Win, Town, Voxel and Judgment Labs under "Trusted by". Blog case studies with named engineers exist for Kernel, Town and Voxel; no source was found for the other five. | [4, 9] |
| Usage claims | "tens of thousands" of changes shipped with Change Monitors; 32% fewer change-related issues for early access customers | C | The first is in Dealroom's item, the second in the launch post ("within two weeks of installing Firetiger"). Neither gives a method or a count of customers. Not verified. | [2, 3] |
| Model spend for three agent types, June 2026 | about US$606,000 a year on Anthropic, about US$231,000 a year on DeepSeek | C | A Firetiger engineer's post, 16 June 2026, which calls the change a 62% reduction. This is a cost, not revenue. Calculated check: (606,000 - 231,000) / 606,000 = 61.9%. | [8] |
| Time from public launch to the sale | 6 months | calc | Calculated: 18 February 2026 to 13 August 2026. | [1, 3] |
How to read this table: a grade A value is stated by the buyer, Cursor, in its own announcement. A grade C value is a statement by Firetiger on its archived pages, a report by Dealroom, or our own request to the site. A "calc" value was computed on this page. No money amount for the sale or for revenue exists in the sources.
| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Bootstrap (plan page, 8 June 2026) | 0 | 300 agent operations a day; 5-day data retention | |
| Growth (plan page, 8 June 2026) | 599 | 300 free agent operations a day plus 25,000 a month; 14-day data retention | |
| Extra agent operations (plan page) | 20 per 600 operations | add-on to Bootstrap and Growth | |
| Enterprise (plan page) | custom | unlimited agent operations; 30-day data retention | |
| Investigation (from 9 June 2026) | 0.50 each | metered | |
| Change Monitor (from 9 June 2026) | 5.00 per pull request watched | plus 0.25 for each extra environment | |
| Custom agent run (from 9 June 2026) | 0.25 per agent session | active services and objectives free |
The company charged for agent work, not for the amount of data sent to it; the plan page said "Send all the data, pay for agents. Storage is on us." The February 2026 launch post put paid plans at US$599 a month with free trials. On 9 June 2026 self-serve plans moved to metered prices per unit of work with a 2-week free trial, and Service Monitors, then in beta, were free. No yearly prices were published, so the yearly column is empty.
When did Firetiger launch, and how did it grow?
From a 2024 start to a sale six months after the public launch, and a shutdown of the product. Each entry carries its grade.
- 2024ARustam Lalkaka, who had spent seven years at Cloudflare, and Achille Roussel, technical lead of Segment's Centrifuge project, start Firetiger. [1, 3]
- 18 February 2026CPublic launch with self-serve plans from US$599 a month and a US$7.6 million seed round led by Sequoia Capital, as the launch post states. [3]
- 22 April 2026CChange Monitors become generally available: an agent per pull request that plans and runs checks after the deploy. [7]
- June 2026COn 9 June self-serve plans move to usage prices per check; on 16 June the company writes that three agent types moved from Anthropic's Claude to DeepSeek models to cut inference cost. [6, 8]
- 13 August 2026ACursor announces that the Firetiger team, with both founders, is joining it. [1]
- 14 August 2026CDealroom reports that terms were not disclosed, new signups are closed, active teams have end dates and the standalone product is being shut down. [2]
- 7 November 2026Cwww.firetiger.com answers HTTP 402 "DEPLOYMENT_DISABLED"; the product is gone. [10]
Why did Firetiger work?
Four causes, in the order they mattered. Each one is a reusable lesson.
Agents that check the work of coding agents
The launch post's argument: coding agents increase the number of changes each day, but the tools that check changes in production were built for people who look at dashboards. Firetiger put an agent on each pull request, let it write a plan from the diff, checked each environment after the deploy, and sent a root cause back to Cursor, Codex or Claude Code through an MCP server or a native integration. The buyer, Cursor, gave the same reason for the purchase.
Pay for agent work, not for data
Observability vendors charge mainly by the amount of data that customers send, which the founders called an adversarial position. Firetiger said storage was free and charged per agent operation, and from June 2026 per investigation and per pull request watched, so its price followed the checks it ran.
Founders from large production systems
The founders had run production at Cloudflare, Segment and Twitch, and the company published long engineering posts on its agent runtime and model costs, plus three customer case studies with named engineers. That record, not a revenue figure, is what the sources show behind the seed round and the sale.
How it was built
These points are the company's own statements on its archived blog and home page, grade C. [3, 4, 7, 8, 9]
- A telemetry store built directly on Amazon S3, which the company called an observability database for the agentic era; a blog post describes S3 and AWS Lambda as the core primitives of its agent runtime.
- A durable execution layer above the database, so that agents could work on goals for weeks or months, call tools, and keep memory between sessions.
- Models: Anthropic's Claude, mostly through AWS Bedrock, for about a year; in June 2026 three agent types moved to DeepSeek v4 Pro with reasoning, after an automated loop rewrote prompts, with an eval score of 92% against 94% for Claude Sonnet 4.6.
- Evals in two sets: static tests on stored data, and a living set that adds interesting agent sessions every week and drops old ones.
- Integrations listed on the home page: GitHub, Linear, OpenTelemetry, Datadog, Postgres, MySQL, ClickHouse, AWS, GCP, Slack, PagerDuty and others; an MCP server let a coding agent start a Change Monitor with monitor_pr.
- The team used Claude Code on its own product, the launch post says: "As Claude Code got more capable, we shipped more."
Tools named in the sources
Each one opens its page in the Tools tab, with every story that uses it.
Who tried this before Firetiger, and why did they fail?
Four companies in our database of failed startups sold monitoring or code-checking tools to software teams and failed, one of them after a sale to a large buyer. The values below are from that database and are not independently verified.
OpTier
The promiseBusiness transaction monitoring and application performance management for large companies.
Why it diedAppDynamics and New Relic moved faster and fitted cloud-native software better.
Verelo
The promiseWebsite monitoring for uptime, malware and performance from many places in the world.
Why it diedIt spent its own budget in six months and could not get businesses to pay for a separate monitoring tool.
Sider
The promiseAutomated code review and static analysis that catch bugs and security flaws before they reach production.
Why it diedFree open-source tools and large security suites that included code analysis left it little room.
Fig DevTool
The promiseAutocomplete and visual command builders for the terminal.
Why it diedFew developers paid in a market of free tools; AWS acquired it in 2024.
The failed startups come from our database of 1,749 dead startups, retold here in our own words. Its entries are AI-assisted summaries of public sources, so these cards carry no evidence grade.
What changed
OpTier and Verelo sold monitoring for people: dashboards and alerts that an engineer had to set up and watch. Firetiger sold agents that wrote their own monitoring plan for each code change and handed the result to another agent, and it charged for the checks, not for the data. Sider checked code before it shipped; Firetiger checked it after the deploy, where real traffic tests code that coding agents wrote.
The end looks like Fig's: a large company took the team and the standalone product stopped. The difference is the buyer's reason. Cursor makes the coding agents whose output Firetiger checked, and it wrote that bringing the two together means "an agent can ship a change, see how it behaves, and respond when something goes wrong."
Can one person copy Firetiger now?
The product, hardly. The narrow idea, yes.
Firetiger had two founders with years of production work at large internet companies, a team, and US$7.6 million of seed money, as the company stated. One person cannot easily copy its data store and agent runtime. The narrow idea is open: check each change that a coding agent makes against what happens in production, and charge per check.
Where it is still open: Smaller checks for one stack or one kind of change: database migrations, mobile releases, payment flows, or one metric such as checkout latency.
- 1Attach an agent to each pull request that writes a short plan of what should change in production.
- 2After the deploy, compare the named metrics with the period before the change, per environment.
- 3Send the finding back to the coding agent through MCP, so that the fix is one pull request away.
- 4Price per check or per investigation, not per gigabyte of data.
What could go wrong for Firetiger?
The risks are the mirror image of the causes above.
What is not verified
No sale price, revenue figure, customer count or headcount was found. We read Cursor's announcement, Dealroom's news item, and Internet Archive copies of Firetiger's home, pricing and blog pages. Firetiger's own post on the sale (blog.firetiger.com/firetiger-is-joining-cursor/) could not be opened: the blog host's security certificate no longer matches its name, and the Internet Archive had no copy. The home page showed eight customer logos under "Trusted by"; blog case studies support three of them (Kernel, Town, Voxel), and we found no source for the other five (Clerk, Avora, Architect, Scale to Win, Judgment Labs). The launch post's 32% fewer change-related issues and Dealroom's "tens of thousands" of changes come with no method or count. A reader could look for a statement by the lead investor, Sequoia Capital, or a later interview with the founders.
The product was shut down after the sale
Dealroom reported on 14 August 2026 that new signups were closed, active teams had service end dates and all user data would be deleted afterward. On 7 November 2026 www.firetiger.com answered HTTP 402 "DEPLOYMENT_DISABLED". Customers had to move to another tool; Cursor said Change Monitors would come to Cursor "soon".
Dependence on model providers
Every agent was a loop around a language model, and the company wrote in June 2026 that inference was its largest cost of goods sold. It ran Anthropic's Claude, mostly through AWS Bedrock, for about a year, then moved three agent types to DeepSeek v4 Pro and stated that their yearly spend fell from about US$606,000 on Anthropic to about US$231,000 on DeepSeek. The post does not say which service hosted DeepSeek or where customer telemetry was processed.
Competition
Firetiger competed with the observability vendors that it criticised, such as Datadog, and on this site Mendral (an AI DevOps agent for CI failures, now closed) and Lucent (AI that watches session replays for bugs) work on nearby problems, while ZeroEntropy is another two-founder AI company whose product was shut down after its team joined a buyer.
Sources
The grade next to each source is the best grade that the source can support. "Read in full" means that the text of the page was downloaded and the numbers were checked in it.
- A Cursor blog, 13 August 2026: Firetiger joins CursorRead in full. The buyer's own announcement, by Maxime Prades. Source for the acquisition, the founders, the 2024 start and Cursor's plans for Change Monitors.
- C Dealroom News, 14 August 2026: Cursor acquires Firetiger to close the loop from code to productionRead in full, in the copy that the pipeline's harvester downloaded on 7 November 2026 (dealroom.co allows ClaudeBot). A news summary. Source for the undisclosed terms, the closed signups, the shutdown of the product and the "tens of thousands" of changes.
- C Firetiger blog, 18 February 2026: Introducing FiretigerRead in full, Internet Archive copy of 10 July 2026. The founder's launch post. Source for the launch, the seed round, the US$599 price, the 32% claim, the founders' background and the use of Claude Code.
- C Firetiger home page, Internet Archive copy of 8 June 2026Read in full. Source for the product description, the logo row, the testimonials and the integrations.
- C Firetiger pricing page, Internet Archive copy of 8 June 2026Read in full. Source for the Bootstrap, Growth and Enterprise plans.
- C Firetiger blog, 9 June 2026: New, simpler pricing for FiretigerRead in full, Internet Archive copy of 10 July 2026. Source for the usage prices and the free trial.
- C Firetiger blog, 22 April 2026: Firetiger Change Monitors: does your PR do what it says on the tin?Read in full, Internet Archive copy of 10 July 2026. Source for how Change Monitors worked and their general availability.
- C Firetiger blog, 16 June 2026: Migrating from Claude to DeepSeek without breaking everythingRead in full, Internet Archive copy of 10 July 2026. An engineer's post. Source for the models, AWS Bedrock, the eval scores and the inference spend.
- C The Firetiger Blog, index page, Internet Archive copy of 10 July 2026Read in full. Source for the list of posts and authors, the three customer case studies and the summary of the S3 and Lambda post.
- C www.firetiger.com, requested 7 November 2026Could not open: the server answered HTTP 402 Payment Required with the body DEPLOYMENT_DISABLED. Source only for the fact that the site is switched off.
- C Database of failed startups, local exportLocal export. AI-written summaries, not independently verified. Source for the failed predecessors.
Run this company? Send corrections or a verified figure to aiolum@livyden.com