# ThinkHive

Aiolum story: https://aiolum.com/story/thinkhive/ · added 2026-11-07 · updated 2026-11-07

**In short.** ThinkHive (Reliability platform that traces and evaluates AI agents in production, United States): Acquired by Rootly, July 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders (Nour Alkhatib and Abdulwahab Omira); headcount at the sale not published. Launched 1 September 2024; sold 22 months after launch.

Reliability platform that traces and evaluates AI agents in production · United States · AI-native · team of 2 to 10 · sold · group: Developer tools and coding agents

- Outcome: Acquired by Rootly, July 2026 (evidence grade of the headline number: E; no revenue figure)
- Launch: 1 September 2024; months from launch to the outcome: 22
- Team: 2 co-founders (Nour Alkhatib and Abdulwahab Omira); headcount at the sale not published
- Outside funding: No outside funding round found in the sources (not verified)
- Main factor: Domain experience
- Meters, 1 to 4: evidence strength 1, copyable pattern 2, capital needed 2
- Tags: AI-native, Small team, Bootstrapped, Exit, B2B, AI infrastructure

Links:

- Visit thinkhive.ai: https://www.thinkhive.ai/

**The product.** A service for engineering teams that run AI agents in production. A small code library sends a record of every step of each agent run to ThinkHive, which checks whether the agent did its job, finds wrong answers and slow decline in quality, groups the failures into patterns, suggests fixes and tests them before they reach users.

**Why it worked.** One of the founders had done this checking by hand every week while she led AI products at Instacart, so the two founders built the tool that she had needed. Rootly, a company that sells incident-management software and ships its own AI agents, bought it in July 2026 to check those agents and to sell agent reliability to its customers.

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

ThinkHive is an observability product for AI agents. Observability means collecting enough records from a running system (logs, metrics and traces, where a trace is the step-by-step record of one request) to explain what it did. ThinkHive adds evaluation, an automatic check of whether each answer was correct, to catch hallucination (an answer the model made up) and drift (quality that falls weeks after launch because the inputs changed). It then tests a fix with shadow testing: the new version runs on real traffic in parallel, and users still get the old answers.

Rootly, the buyer, says Nour Alkhatib and Abdulwahab Omira founded ThinkHive after Alkhatib had spent weekly sessions at Instacart working out why its agents failed without any alert. ThinkHive's changelog dates its first public release to 1 September 2024, and its terms of February 2026 describe paid subscription plans billed through Stripe. Rootly announced the purchase on 28 July 2026, and the team joined Rootly to lead its work on agent reliability.

No price for the sale, no revenue and no count of paying customers appear in any source. This is a grade E story: the sale is confirmed by the buyer, but the business's size is not known.

## Pricing

ThinkHive plans as the billing screen of its app names them, read from the app's code on 7 November 2026 (grade C); the app loads the amounts from its server after login, so no price was read

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Free | 0 | 0 | credit-limited; the app labels it "Free Forever" |
| Pro | not published | not published | monthly or annual billing; the app shows "Save ~24%" for annual |
| Enterprise | contract | contract | "Contact Sales"; a service level agreement may apply |
| Credit packs |  | not published | extra credits bought in the app, priced per credit |

The terms of 20 February 2026 say the service is sold as subscription plans with credit-based usage metering and that Stripe processes all payments. The app also sells extra credit packs, with a price per credit loaded from the server. A free tier is offered at the company's discretion. Enterprise customers may get a service level agreement. We found no public pricing page.

## How much did ThinkHive make, and what was it sold for? The numbers, graded

A two-founder AI agent reliability product, sold to Rootly in July 2026 for a price that was not disclosed

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Outcome | Acquired by Rootly, announced 28 July 2026 | A | Rootly's chief executive wrote the announcement. Dealroom reported the deal the same day. | [1, 2] |
| Price of the sale | not disclosed | B | Dealroom: "Terms were not disclosed." Rootly's announcement gives no amount. | [2, 1] |
| Founders | Nour Alkhatib (chief executive) and Abdulwahab Omira | A | Rootly: "ThinkHive was founded by Nour Alkhatib and Abdulwahab Omira". Alkhatib had led AI products at Instacart. | [1, 2] |
| Team at the sale | headcount not published | A | Rootly writes that "the team that built it is joining Rootly" to lead its work on agent reliability, without a number. No source suggests more than 10 people. | [1] |
| First public release | 1 September 2024, by the company's changelog | C | Version 1.0.0, "First public release of ThinkHive". The changelog's 2024 dates may be a year early: it dates version 3.0.0 of the SDK to 30 December 2024, but the npm registry shows 3.0.0 first published on 29 December 2025. | [5, 7] |
| Paid plans | Free, Pro and Enterprise, plus credit packs | C | Named in the billing screen of the app, with "Upgrade to Pro" and "Contact Sales" buttons and Stripe checkout. The terms describe recurring subscription fees charged by Stripe. Prices not read. | [6, 4] |
| Usage billing | credit system with Stripe, added in version 2.0.0 | C | Changelog entry dated 1 October 2024: "Credit System: Usage-based billing with Stripe integration" and "Rate limiting by tier". | [5] |
| JavaScript SDK on npm | 16 versions from 29 December 2025 to 22 March 2026 | C | Package @thinkhive/sdk, MIT license, maintainer account abdul-omira. An SDK (software development kit) is the code library that customers add to their agent to send traces. | [7] |
| Named teams on the home page | nsave, aiXplain and Caleem under "Trusted by teams at" | C | Not described as paying customers and not linked to case studies. A second row, "Built with input from top AI teams at 20+ companies", shows Google, Spotify, FluidAI, Homebase and Infios. See "Trust signals" below. | [3] |
| Company | ThinkHive, Inc. | C | The terms choose the law and the courts of Delaware, United States. The headquarters city is not stated. | [4] |
| Time from first release to the sale | 22 months | calc | Calculated: 1 September 2024 to 28 July 2026. If the changelog's 2024 dates are a year early, it was about 10 months. | [5, 1] |

How to read this table: a grade A value is stated by the buyer, Rootly, in its own announcement. A grade B value is from a news item by Dealroom. A grade C value is a statement on ThinkHive's own pages or a public registry record. A "calc" value was computed on this page. No money amount for the sale or for revenue exists in the sources.

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

From hand-made checks of AI agents at Instacart to a sale to an incident-management company, in about 22 months by the company's own release dates. Each entry carries its grade.

- **Before ThinkHive** (grade A [1]): Alkhatib leads AI products at Instacart, where agents serve millions of customers. Every week she and her team do forensic work to understand why the agents do not improve business results. That effort leads her to leave and start ThinkHive with Omira.
- **1 September 2024** (grade C [5]): Version 1.0.0, the first public release by the company's changelog: trace capture through OpenTelemetry, failure clustering, fix suggestions with AI, and SDKs for JavaScript and Python. A month later, version 2.0.0 adds credit-based billing through Stripe.
- **29 December 2025** (grade C [7]): The JavaScript SDK @thinkhive/sdk is first published on npm, as versions 2.0.0 and 3.0.0 on the same day.
- **February to March 2026** (grade C [5, 4]): Version 4.0.0 links agent quality to business metrics and return on investment. The terms are updated on 20 February. Version 4.1.0 adds a Guardrails API that scans content for personal data and secrets in real time.
- **28 July 2026** (grade A [1, 2]): Rootly announces that it has acquired ThinkHive. Terms are not disclosed. Rootly's own agents will run on ThinkHive's evaluation engine, and the team joins Rootly.
- **7 November 2026** (grade C [3]): The ThinkHive site is still online with the banner "as of July 2026, ThinkHive is part of Rootly" and a "Start for Free" button.

## Why did ThinkHive work?

### A problem the founder had solved by hand

Rootly's announcement says the idea came from weekly manual investigations at Instacart into agents that failed without any alert. The founders built the product for a task they had already done themselves.

### Several signals instead of one score

ThinkHive combines metrics, traces and evaluations, and it has rule-based checks that do not use a model at all. Alkhatib said, in a quotation that Dealroom reports: "Using AI to judge AI is like asking the same student to mark their own exam." Rootly named this approach as a reason for the purchase.

### A buyer that needed it for itself

Rootly ships AI agents into its customers' incident response. It bought ThinkHive both to check its own agents and to sell agent reliability to customers whose agents now cause incidents of their own.

## How it was built

These details come from ThinkHive's documentation changelog, its site and terms, and the npm registry. They are grade C.

- Customers add an SDK for JavaScript or Python, or send traces through OpenTelemetry, an open standard for tracing data. The site says it works with agents built on OpenAI, Anthropic, LangChain or in-house code.
- Analysis features in the changelog: hallucination detection of 9 types, groundedness and citation scores for answers that use retrieved documents, a "Jury Mode" that combines the votes of several models, and rule-based graders for length, keywords, JSON validity and personal data.
- An MCP server (the Model Context Protocol lets an AI assistant call outside tools) with more than 25 tools for Claude Code, by the changelog.
- Login through Auth0 with single sign-on, redaction of personal data with 37 detection patterns, and audit logs.
- Payments through Stripe, with use measured in credits ("credit-based usage metering" in the terms).

Sources: [5, 3, 4, 7]

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

Two companies in our database of failed startups promised to find failures in running software and tie them to business results, and lost to larger competitors. 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.

### Proxino (2011 to 2014; funding US$120K; cause: Competition)

- Promise: Real-time detection of JavaScript errors in users' browsers, to make debugging faster.
- Why it died: Sentry and other tools offered more complete products. It did not build a developer community, had reliability problems in its own analytics, and ran short of money.
- Record: 

### OpTier (2003 to 2015; funding US$112M; cause: Competition)

- Promise: Monitor billions of IT and business transactions a day and turn them into business insight.
- Why it died: AppDynamics and New Relic moved faster, and it did not adapt to cloud-native systems.
- Record: 

## What changed

Both predecessors sold monitoring as a separate product and lost when larger monitoring companies covered the same need. The same risk was open for ThinkHive, and the outcome was different: it did not try to outgrow the large vendors but was bought by an incident-management company that needed exactly this feature for its own AI agents.

The failure it catches is also newer. A crashed server sets off every alarm, but an agent that gives a wrong answer with normal speed sets off none, so the older monitoring tools do not see it. A small team that knew this failure from inside a large company could build the checks from first-hand knowledge.

## Can one person copy ThinkHive now?

**Verdict: The product, with difficulty. The path to a sale, yes.**

A full agent-evaluation platform with SDKs, single sign-on and compliance features is a lot of work for one person, and the company's own comparison table lists two groups of competing tools. The path is open: solve a narrow failure that you have seen inside a large company, build it as a feature that a larger software company needs for its own AI, and that company may buy it.

1. Write down a check that you or your team do by hand every week on an AI system.
2. Turn it into a library and a dashboard that other teams can add in minutes.
3. Combine model judges with rule-based checks, so the result does not rest on one model's opinion.
4. List the software companies that ship AI agents to their own customers: they are both the buyers and the possible acquirers.

**What is still open.** Checks for one kind of agent at a time: voice agents, coding agents, support agents, and agents that act inside a single business system.

## What could go wrong for ThinkHive?

- **What is not verified.** No sale price, revenue figure or count of paying customers was found. We read Rootly's announcement, Dealroom's news item, ThinkHive's home page, terms and privacy policy, its documentation changelog, the billing screen of its app and the npm registry. Business Wire's server did not answer our request for its robots.txt, so Rootly's press release there was not read; PitchBook answered 403, PyPI showed a bot challenge and the Internet Archive answered 429, so those were not read. The launch date rests on the changelog, whose 2024 dates disagree with the npm registry by a year. No funding round was found, but Dealroom's full profile is behind a paywall, so outside funding is not ruled out. The site still offers sign-up after the sale; whether Rootly keeps selling ThinkHive on its own is not stated. A reader could look for the Business Wire release or an interview with the founders.
- **Trust signals on the home page.** A row headed "Trusted by teams at" shows nsave, aiXplain and Caleem without saying that they are customers and without a case study. A second row, "Built with input from top AI teams at 20+ companies", shows the logos of Google, Spotify, FluidAI, Homebase and Infios; the page's screen-reader text calls them "Companies whose AI teams gave input". A row headed "Supported by leaders in business" shows Microsoft for Startups, NVIDIA, Perplexity and Anthropic, without saying what the support was. No source that we read confirms any of these relations.
- **Competition.** ThinkHive's own comparison table places it against "Generic Observability" and "LLM Tracing & Eval Tools" without naming them. On this site, Agnost AI (monitoring for AI agents) and Kyvvu (a check on each action of an AI agent before it runs) sell to the same engineering teams.
- **Dependence on model providers.** The changelog lists "Fix suggestions with AI" and a "Jury Mode" that combines several models, but no source names which providers' models ThinkHive itself calls. Its rule-based graders do not need a model. The OpenAI and Anthropic names on the site refer to the customers' agents that it can trace.

## Sources

1. Rootly blog, 28 July 2026: Why Rootly acquired ThinkHive — https://rootly.com/blog/why-rootly-acquired-thinkhive (grade A). Read in full. The buyer's own announcement, by its chief executive. Source for the sale, the founders, the Instacart background and the team joining Rootly.
2. Dealroom News, 28 July 2026: Rootly buys ThinkHive to make its AI incident agents provable — https://dealroom.co/news/141613-rootly-buys-thinkhive-to-make-its-ai-incident-agents-provable/ (grade B). Read in full (downloaded by the pipeline on 7 November 2026). Source for "Terms were not disclosed" and the quotations of Rootly's chief executive and Alkhatib.
3. ThinkHive home page, read 7 November 2026 — https://www.thinkhive.ai/ (grade C). The page is built by script, so its text was read in the site's script file, downloaded 7 November 2026. Source for the acquisition banner, the logo rows, the sign-up buttons and the supported agent frameworks.
4. ThinkHive Terms of Service and Privacy Policy, last updated 20 February 2026 — https://www.thinkhive.ai/terms (grade C). Read the sections on plans, payment, the free tier and governing law, in the site's script file. Source for the subscription plans, Stripe, ThinkHive, Inc. and Delaware law.
5. ThinkHive Docs: Changelog — https://docs.thinkhive.ai/changelog/ (grade C). Read in full. Source for the release dates from version 1.0.0 to 4.1.0 and the features of each version.
6. ThinkHive app: billing screen code, read 7 November 2026 — https://app.thinkhive.ai/ (grade C). Read the text of the billing and checkout script files. Source for the Free, Pro and Enterprise plans, credit packs and Stripe checkout. The amounts load from the server and were not read.
7. npm registry record for @thinkhive/sdk — https://registry.npmjs.org/@thinkhive%2Fsdk (grade C). Read the version dates, the maintainer and the description. Source for the first publication on 29 December 2025 and the last version on 22 March 2026.
8. 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 ThinkHive? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: ThinkHive).