Aiolum Browse stories
All stories
Multi-agent workspace on a visual canvas for research, decks and models · United States

Spine AI: Product, team and how it was built

Small teamFundedB2BBusiness software

Spine AI (Multi-agent workspace on a visual canvas for research, decks and models, United States): Live with paid plans from March 2026, November 2026. Evidence grade E: no revenue figure was checked. Team: 2 founders in 2023 and at the launch post of 13 March 2026; 7 people on the Y Combinator profile read in November 2026. Launched April 2023; still running.Added

Outcome
Live with paid plans from March 2026, November 2026
Evidence
E no revenue figure was checked
Launch
April 2023
Status
Still running
Team
2 founders in 2023 and at the launch post of 13 March 2026; 7 people on the Y Combinator profile read in November 2026
Outside funding
Y Combinator Summer 2023; amount not found
Main channel
Not stated
No revenue figure

What is not verified

No revenue figure was found. The pages read were the company's home, pricing, founder and consultant pages, a benchmark post, the Launch HN thread and the Y Combinator profile. Also read were archived copies of 2023 and 2024. No payment-processor profile, filing or press piece with a figure was found. The company's terms and privacy pages load by script and could not be read, so no billing terms or payment processor were seen. A commenter wrote that they paid one month, but no source from the company or the press gives a customer count. 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.

Spine AI sells a multi-agent workspace. A multi-agent system is a program in which several AI agents, each with its own task, work on one goal. The agents run on a canvas, a large visual board where each step is a block that the user can see and change. The company's pages say it is meant for non-coding work such as market research, financial models and pitch decks.

The company is small and began in 2023. Y Combinator names Akshay Budhkar and Ashwin Venkatesh Raman as its founders and lists a team of 7. Archived pages show an earlier chat product of 2 April 2023 and an AI copilot product of 4 January 2024. The Launch HN post of 13 March 2026 introduced the canvas and agent version.

No revenue figure was found, so this story has grade E. The company's pricing page lists Free, Starter and Max plans and a custom Enterprise plan. A Hacker News commenter wrote that they paid one month. No source gives a customer count or a revenue number.

Spine AI: Live with paid plans from March 2026, November 2026. Multi-agent workspace on a visual canvas for research, decks and models, United States, team of 2 to 10, evidence grade E.
The story in one card. Share it with the address of this page.
01

Why did Spine AI work?

Four causes, in the order they mattered. Each one is a reusable lesson.

01

The founders changed the product several times over three years

The archived pages show a chat over documents in April 2023 and a copilot for other companies' apps in January 2024. The March 2026 post says they spent about three years on many product iterations. Y Combinator lists further earlier launches without dates.

02

Users who did not want to learn the canvas asked for chat

The launch post says the first canvas was fully manual and that new users kept asking for a chat layer. The founders wrote that building it showed them that agents could run for hours and produce complete deliverables.

03

It launched in public with a free tier and credit plans

The launch post links the pricing page and says there is a free tier. It also says agents use more credits than manual work. A commenter who ran out of free credits wrote that they paid for one month to see the result.

How it was built

These details come from the launch post, the company's own pages and its benchmark post. They are grade C. [4, 8]

  • Each unit of work is a block: a language model call, image generation, web browsing, an app, a slide deck or a spreadsheet. Any block can connect to any other and passes its context on.
  • A central orchestrator splits a task into subtasks and gives each to a specialised agent. Agents may change the model and the prompt of a block.
  • The launch post says the system is model-agnostic. It names OpenAI language models, the image model Nano Banana Pro and Claude as examples.
  • Agents can stop and ask the user a question before they go on. The launch post says this keeps a person in the loop.
  • A founder wrote in a reply that a demo task cost about 7,000 credits and ran around 10 BrowserUse blocks.

Tools named in the sources

Each one opens its page in the Tools tab, with every story that uses it.

02

Who tried this before Spine AI, and why did they fail?

Three companies in our database of failed startups promised to have AI do skilled office work for customers. The values below are from that database and are not independently verified.

Died 2012 to 2023Burned US$852.0M burned

Olive AI

The promise

An AI workforce of software bots that would handle prior authorizations, claims and billing checks in hospitals.

Why it died

The record says the bots broke often, each customer needed custom work, and the cost of serving a customer was higher than its revenue.

Promise larger than the product
Died 2015 to 2018Burned US$13.0M burned

ROSS Intelligence

The promise

An AI research platform that finds relevant case law for lawyers.

Why it died

The record says established legal databases held the data and the customers, and lawyers did not see the AI as a full replacement.

Market too hard to enter
Died 2016 to 2025Burned US$445.0M burned

Builder.ai

The promise

Custom software built through an AI-assisted assembly line, so people without developers could get an app.

Why it died

The record says the AI was mostly a layer over a service business, and delivery cost more than the projects earned.

Unit economics

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

All three records sold finished work that an AI was said to produce. Olive and Builder.ai relied on large teams of people behind the AI, and their costs grew with each customer. ROSS met rivals that owned the data. The records give these as the reasons they failed.

Spine AI sells credits that are used when agents run, so its price rises with the work that each customer asks for. The company's own launch post says agents use more credits than manual work. No source here shows what a task costs the company or what margin it earns.

03

Can one person copy Spine AI now?

Not yet. A small team ran it for three years, and no source shows that customers pay enough.

The method is visible: build on public models, charge by credits, and keep changing the product until users stay. This page cannot show that it earns money. No revenue figure, customer count or cost figure was found, so a reader has no evidence about margins.

Where it is still open: Other kinds of long AI work that users now do by hand: audits, grant applications, due diligence and technical documentation.

  1. 1Start with one narrow product on public models and see what users keep asking for.
  2. 2Charge by credits so that heavy use pays for the model cost it creates.
  3. 3Launch in public with a free tier that lets a user finish one real task.
  4. 4Publish a revenue figure from a payment processor so that readers can check it.
04

How much revenue does Spine AI make?

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.

No revenue figure is public. The facts below show a price list, one commenter who paid, a start in 2023 and a team of seven
FactValueGradeHow we knowSources
Plans and pricesFree, Starter and Max plans; Enterprise on requestCThe company's pricing page, read on 12 November 2026, lists Starter from US$16 a month billed yearly. It lists Max from US$80 a month billed yearly.[2]
A user's account of payingone commenter paid the one-month fee in March 2026CA Hacker News commenter wrote on 13 March 2026 that they hit the token cap on the free tier and paid for one month. Their words were "I decided to gamble the one month fee to let it continue". They wrote that the payment form defaulted to annual billing.[4]
Credit top-up10,000 credits for US$10, one-timeCThe company's pricing page says the credits never expire. It also says top-ups are for Pro and Max subscribers, although the table names Starter and Max.[2]
Launch of the canvas and agent product13 March 2026, Hacker NewsCThe post is titled "Launch HN: Spine Swarm (YC S23)". It says the pricing is usage-based credits and that there is a free tier.[4]
Earlier products of the same companychat over documents, April 2023; AI copilot, January 2024CArchived copies of getspine.ai of 2 April 2023 and 4 January 2024 show these two products. Y Combinator lists other launches under the same company name.[6, 7, 5]
Company start2023CY Combinator lists "Founded: 2023" and "Batch: Summer 2023". The archived pages of 2023 and 2024 carry a copyright line for "Spine AI" in 2023.[5, 6, 7]
FoundersAkshay Budhkar, founder and chief executive; Ashwin Venkatesh Raman, founder and chief technology officerCThe Y Combinator profile names them. The launch post says "We're Ashwin and Akshay" and "friends for over 13 years".[5, 4]
Team size7 on the Y Combinator profile, read in November 2026CThe profile lists "Team Size: 7" with no date. The launch post of March 2026 names two founders and gives no headcount.[5, 4]
FundingY Combinator Summer 2023; amount not foundCThe profile lists the batch. No page read states a funding amount or an investor other than Y Combinator.[5]
Time buildingabout 3 yearsCThe launch post says the founders have "spent about 3 years building Spine across many product iterations".[4]
Benchmark claim87.6% on DeepSearchQA, February 2026CThe company states this score for its own system on a Google DeepMind benchmark. It lists lower scores for four rivals. No independent check was read.[3, 4, 8]
Time from first page to the headline43 monthscalcCalculated: April 2023 to November 2026.[6, 2]

How to read this table. Grade C is a statement by the company, its founders or a profile page, or a comment by a user. A grade E story has no dated money figure, so no row here is a revenue number.

Spine prices as listed on the company's pricing page on 12 November 2026 (grade C)
PlanPer month, billed yearly (US$)Per month, billed monthly (US$)Limit
Free002,000 credits a month
Starter, 30,000 credits16201,500 daily refresh credits
Starter, 50,000 credits24301,750 daily refresh credits
Starter, 75,000 credits32402,000 daily refresh credits
Starter, 100,000 credits40502,250 daily refresh credits
Max, 200,000 credits801005,000 daily refresh credits
Max, 350,000 credits1201496,000 daily refresh credits
Max, 500,000 credits1591997,500 daily refresh credits
Max, 750,000 credits23929910,000 daily refresh credits
Enterprisecustom credits, SSO and support

The company's pricing page sells credits, which the agents use when they run blocks. Plans renew monthly or yearly, and the credit tiers are set with a slider. The page also lists a one-time top-up of 10,000 credits for US$10, and a student discount on request by email. The page names Starter and Max in its price table, while its FAQ speaks of Pro and Max subscribers.

05

When did Spine AI launch, and how did it grow?

From a chat over documents in April 2023 to a canvas product with paid plans in November 2026. Each entry carries its grade.

  1. 2 April 2023CAn archived copy of getspine.ai shows Spine as a way to build a chat interface over a customer's own data. It lists a free plan and a premium plan marked "Coming soon!". It names two founders. [6]
  2. Summer 2023CY Combinator lists the company in its Summer 2023 batch and as founded in 2023. [5]
  3. 4 January 2024CAn archived copy shows a new product: an AI copilot that a software company adds to its own app through the app's APIs. The page says "Backed by Y Combinator". [7]
  4. 25 February 2026CThe company publishes a post on its benchmark results for Spine Swarm. Four authors are named on the post. [8]
  5. 13 March 2026CThe founders launch Spine Swarm on Hacker News with a free tier and credit-based paid plans. A commenter writes the same day that they paid one month. [4]
  6. 12 November 2026CThe company's pages list Free, Starter and Max plans, a canvas product and a local product called Medley. Y Combinator shows a team of 7. [1, 2, 5]

What could go wrong for Spine AI?

The risks are the mirror image of the causes above.

What is not verified

No revenue figure was found. The pages read were the company's home, pricing, founder and consultant pages, a benchmark post, the Launch HN thread and the Y Combinator profile. Also read were archived copies of 2023 and 2024. No payment-processor profile, filing or press piece with a figure was found. The company's terms and privacy pages load by script and could not be read, so no billing terms or payment processor were seen. A commenter wrote that they paid one month, but no source from the company or the press gives a customer count.

Earlier products and the start of the company

The company sold a chat over documents in April 2023 and an AI copilot for other software in January 2024. Y Combinator lists the company as founded in 2023, so the start is 2023. The canvas and agent product that this story describes launched in March 2026. A reader should treat the 2023 start as the company's, not the current product's.

Benchmark claims and badges

The home page says the product ranked first on two benchmarks in February 2026 and gives new highs for Medley in July 2026. It says Canvas is SOC 2 Type I compliant. These are the company's statements, and no report or independent check was read. The pages read show no logo wall, counter or quotation from a customer. The archived page of 4 January 2024 carried two named quotations and the line "Join other teams already adding up to $20M to their ARR with AI Agents" with no source. Neither quotation is from a named customer. One is by the person whom Y Combinator lists as the company's primary partner, and the page did not say so.

Use cases and pricing wording

The home page lists example prompts that research people and companies from public sources. Examples are a judge's rulings, an opposing lawyer's win rate, and a prospect's LinkedIn background and recent posts. It also lists prompts for outreach drafts. The product does not sell contact data. The pricing page offers a student discount to university students, and its FAQ names Pro and Max plans while its table names Starter and Max.

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.

  1. C Spine AI home page
    Read in full, 12 November 2026. Source for the two products, the benchmark and SOC 2 claims and the example prompts.
  2. C Spine pricing page
    Read in full, 12 November 2026. Source for the plans, the credit tiers, the top-up and the FAQ.
  3. C Spine for founders
    Read in full, 12 November 2026. Source for the benchmark scores that the company lists.
  4. C Launch HN: Spine Swarm (YC S23) – AI agents that collaborate on a visual canvas, Hacker News, 13 March 2026
    Read in full, with the comments, 12 November 2026. Founder post. Source for the launch, the three years, the pricing note and a commenter who paid.
  5. C Spine AI at Y Combinator
    Read in full, 12 November 2026. Source for the founders, the start year, the batch, the team size and the list of earlier launches.
  6. C Internet Archive copy of getspine.ai, 2 April 2023
    Read in full, 12 November 2026. The archive named it as the closest copy to 1 April 2023. Source for the first product and the 2023 copyright line.
  7. C Internet Archive copy of getspine.ai, 4 January 2024
    Read in full, 12 November 2026. The archive named it as the closest copy to 1 January 2024. Source for the copilot product and the old claims.
  8. C Spine Swarm Hits #1 on GAIA Level 3 and Google DeepMind DeepSearchQA, 25 February 2026
    Read the first sections, 12 November 2026. Company post. Source for the architecture and the authors.
  9. C getspine.ai sitemap, with the terms of service and privacy pages
    Read in full, 12 November 2026. It lists the terms and privacy pages. Both pages came back as the script shell of the home page, so they could not be read.
  10. C Database of failed startups, local export
    Record pages read on the website on 12 November 2026. AI-written summaries, not independently verified. Source for the failed predecessors.

Run this company? Send corrections or a verified figure to aiolum@livyden.com

Back to all stories