# Twill.ai Aiolum story: https://aiolum.com/story/twill-ai/ · added 2026-11-12 · updated 2026-11-12 **In short.** Twill.ai (Cloud sandboxes that run AI coding agents and return pull requests, United States): Paid plans live, November 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders named in the launch post (April 2026); later headcount not published. Launched April 2026; still running. Cloud sandboxes that run AI coding agents and return pull requests · United States · AI-native · team of 2 to 10 · still running · group: Developer tools and coding agents - Outcome: Paid plans live, November 2026 (evidence grade of the headline number: E; no revenue figure) - Launch: April 2026; months from launch to the outcome: 7 - Team: 2 co-founders named in the launch post (April 2026); later headcount not published - Outside funding: Y Combinator S25; amount not published - Main factor: Product - Meters, 1 to 4: evidence strength 1, copyable pattern 2, capital needed 2 - Tags: AI-native, Small team, Funded, Running, B2B, Software Links: - Visit twill.ai: https://twill.ai/ **The product.** A service that runs AI coding agents such as Claude Code and Codex in cloud sandboxes. A team hands a task over from Slack, GitHub, Linear or a web app. The agent works in its own copy of the project and returns a pull request. **Why it worked.** Its stated difference is that it reuses the coding tools that the AI labs ship, instead of building its own agent. The founders say every improvement from a lab then reaches Twill users. They also say a user can pick a different agent for each task. 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 Twill runs coding agents in the cloud. A coding agent is a program, such as Claude Code or Codex, that writes and changes software code from a written task. The founders say Twill starts a separate sandbox for each task, an isolated cloud computer where the agent runs code. The result comes back as a pull request, a proposed code change that the team reviews before it is merged. The company's pricing page, read on 12 November 2026, lists Pro at US$50 a month and Max at US$200 a month for each cloud workspace. In a launch post dated 10 April 2026, the co-founders Willy and Dan say they are part of Y Combinator's S25 batch. The terms of 12 March 2026 name Anybase Inc., a Delaware corporation, as the operator. An archived copy of the terms, effective 15 August 2025, names Chainlit SAS, a French company that the company register dates to 14 April 2023. No revenue figure was found, so this story is grade E. The headline is a status fact: paid plans are listed on the company's own pricing page. The terms say paid subscriptions are billed through Stripe. No source states that a customer has paid. ## Why did Twill.ai work? ### It runs the labs' own coding tools in a sandbox for each task The founders say Twill does not build its own agent. It starts a sandbox, copies the project, installs dependencies and runs the coding tool the user chose. The post says each task gets its own files, ports and processes, so two tasks do not collide. ### It sells model credits at cost in a plan for each workspace The pricing page says one credit is US$1 of model usage billed through Twill, at cost. Plans are priced for each cloud workspace and not for each seat. The newsletter of 16 March 2026 says users who bring their own keys are exempt from credit billing. ### It takes tasks from Slack, Linear and GitHub and returns pull requests The home page says anyone on a team can start a cloud agent from a message, an issue or a task. The agent pulls in the issue, thread and documents behind the task. It then opens pull requests and posts updates where the team already works. ## How it was built These details come from the company's launch post, privacy policy and repository. They are grade C. - The launch post says Twill reuses the command-line tools of the AI labs, such as Claude Code and Codex. It runs each one in its own sandbox. - The privacy policy names Modal and Daytona as providers of the isolated sandboxes, and Vercel for application hosting. - The privacy policy names Anthropic and OpenAI as providers of AI models, PostHog for usage analytics and Sentry for error monitoring. - The privacy policy names Stripe for payments and Resend for transactional email. - The launch post says Twill snapshots the sandbox after a task so the next run on the same repository starts with dependencies installed. - The company published agentbox-sdk, a TypeScript SDK under the MIT licence, for running agent tools in sandboxes. Sources: [3, 5, 11] ## Who tried this before Twill.ai, and why did they fail? Three companies in a database of failed startups sold developer tools with a similar promise: less setup work for programmers. The values below are from that database, read on 12 November 2026, 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. ### Kite DevTool (2014 to 2024; funding US$21.0M; cause: Timing and monetization) - Promise: AI code completion inside the programmer's editor, with a free tier and a paid Pro version. - Why it died: The record says Kite came about seven years before GitHub Copilot and made missteps in how it charged. - Record: ### Nitrous.io (2012 to 2016; funding US$8.0M; cause: Competition) - Promise: Ready-made cloud development environments, so programmers could code from any machine without local setup. - Why it died: The record says it was squeezed between AWS and another force, with right idea but wrong timing and strong competition. - Record: ### CodeSee (2020 to 2024; funding US$10.0M; cause: Free users did not pay) - Promise: Automatic maps of a codebase, so new team members understand it faster. - Why it died: The record says it struggled to turn free users into paying customers and shut down in 2024. - Record: ## What changed Each of the three sold a tool to programmers and met a larger vendor or free users who did not pay. Twill sells into the same market. The founders themselves say in the launch thread that the labs will keep improving their own agents. Two things differ in what the sources show. Twill runs the labs' own tools instead of competing with them, and it sells model usage at cost. The company's own pages do not say what margin the plans leave. ## Can one person copy Twill.ai now? **Verdict: The structure, yes. The proof of sales, not yet.** The pattern is a thin service around tools that other companies keep improving. No source shows how much Twill earns or how many people work there. 1. Reuse the command-line tools that the AI labs ship, instead of writing your own agent. 2. Give each task its own sandbox, so tasks cannot collide. 3. Sell model usage as credits at cost, in a plan for each workspace. 4. Launch on Hacker News with a free tier and an open-source SDK. **What is still open.** Other jobs where a team wants an AI agent to work unattended: code review, test repair, documentation and routine upgrades. ## How much revenue does Twill.ai make? The numbers, graded A cloud service for AI coding agents with a published price list and a Y Combinator launch; no revenue figure is published | Fact | Value | Grade | Note | Sources | |---|---|---|---|---| | Prices, November 2026 | Pro US$50 a month and Max US$200 a month, per cloud workspace | C | Company pricing page: "$50 / month" with "50 monthly cloud credits" and "$200 / month" with "200 monthly cloud credits". | [1] | | Prices in an archived copy, 9 September 2025 | Pro US$30 a month, with a waitlist button | C | Internet Archive copy of the company's home page. The page said "Join Waitlist" and listed Hobby, Pro and Team plans. | [7] | | Billing model, 16 March 2026 | monthly credits that offset model usage | C | Company newsletter: "We moved from per-run pricing to token credits. Your subscription now grants monthly USD credits." | [6] | | Free tier at launch, 10 April 2026 | 10 credits a month, against 5 on the page of November 2026 | C | Launch post: "Free tier: 10 credits per month". The current pricing page says "5 monthly cloud credits". No source explains the change. | [3, 1] | | Public launch | Launch HN post, 10 April 2026 | C | Hacker News dates the post 2026-04-10. A Startup Fortune article of 11 April 2026 retells the post. | [3, 9] | | Accelerator | Y Combinator, S25 batch | C | The post title says "YC S25". The company's structured page data says "Backed by Y Combinator". No source gives an amount. | [3, 2] | | Founders | Willy and Dan, co-founders | C | Launch post: "we're Willy and Dan, co-founders of Twill.ai". Full names are not stated on a page that ties them to the business. | [3] | | Legal operator, March 2026 | Anybase Inc., a Delaware corporation | C | Terms effective 12 March 2026: "operated by Anybase Inc., a Delaware corporation doing business as Twill". No incorporation date was found. | [4, 5] | | Older company that first operated Twill | Chainlit SAS, France, registered on 14 April 2023 | B | French company register: creation date 2023-04-14, seat in Paris, status active. Archived terms effective 15 August 2025: "operated by Chainlit SAS". It first made Chainlit, an open-source Python package; GitHub dates that repository to 14 March 2023. | [13, 8, 14] | | Payment processing | Stripe, with a customer ID linked to each workspace | C | Privacy policy: "Stripe customer ID linked to your workspace" and "Stripe: Payment processing, subscription management, and billing". | [5] | | Open-source code | agentbox-sdk, MIT licence, 103 commits | C | GitHub repository of the company, read on 12 November 2026. The launch post calls it an SDK for running agent tools across sandbox providers. | [11, 3] | | Time from the launch post to November 2026 | 7 months | calc | Calculated: April 2026 to November 2026, in whole months. | [3] | How to read this table: a grade B value comes from the French company register. A grade C value is what the company says on its own pages. It can also be what an archived copy of those pages shows. A grade X value was computed on this page. No value comes from a payment processor, a filing or a journalist. ## Pricing Twill plans on the company's pricing page, read 12 November 2026 (grade C) | Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit | |---|---|---|---| | Free | | 0 | 5 monthly cloud credits; 1 cloud workspace; up to 2 users | | Pro | | 50 | 50 monthly cloud credits; unlimited users; automations | | Max | | 200 | 200 monthly cloud credits; bring your own API keys; larger machines | | Enterprise | | | contact the company; private cloud, SSO and audit trail | The page shows one monthly price for each plan, per cloud workspace and not per seat, and no yearly price. The company says one credit is US$1 of model usage billed through Twill, at cost. Enterprise has no published price. ## When did Twill.ai launch, and how did it grow? Timeline From terms and a waitlist page in the second half of 2025 to a price list with two paid plans in November 2026. Each entry carries its grade. - **15 August 2025** (grade C [8]): The terms of service carry this effective date and name Chainlit SAS as the operator. They describe an AI project management platform that sends work to coding agents. - **9 September 2025** (grade C [7]): An archived home page shows a waitlist and three plans: Hobby, Pro and Team. - **12 March 2026** (grade C [4, 5]): The terms and the privacy policy name Anybase Inc., a Delaware corporation, as the operator. - **16 March 2026** (grade C [6]): A newsletter says billing moved from per-run pricing to token credits that come with the subscription. - **10 April 2026** (grade C [3]): The founders post a Launch HN thread about running coding agents in cloud sandboxes. - **12 November 2026** (grade C [1]): The pricing page lists Free, Pro and Max plans and an Enterprise contact option. ## What could go wrong for Twill.ai? - **What is not verified.** No revenue figure was found. What was looked for: a payment-processor profile, a press piece with customers, a founder interview, a filing. A TrustMRR profile request answered with a server error. The pricing page has buttons for the paid plans, and the terms say paid subscriptions are billed through Stripe. No source states that a customer has paid. - **Logos and badges on the home page.** The home page shows a "Backed by" badge with the Y Combinator logo. The launch post confirms the S25 batch, and no amount is stated. A row of logos shows Slack, GitHub, Linear, Notion, Asana, Sentry, Datadog, AWS and Google Cloud. These are integrations that the product connects to, and no source says they are customers. - **Start date and operator.** Twill is counted here from its own launch, although an older company first operated it. The French company register dates Chainlit SAS to 14 April 2023. That company first made Chainlit, an open-source Python package for chat applications, and Literal AI, a tool for monitoring AI applications. No page states when Anybase Inc. was founded. The earliest dated page for Twill is the terms effective 15 August 2025. No source says how the two companies relate or when Twill moved from one to the other. Archived copies of twill.ai from before September 2025 that were found are pages offering the domain for sale. - **Team and plan changes.** No page counts the people at the company. The launch post names two co-founders. The free tier shows 10 credits in the launch post of April 2026 and 5 credits on the pricing page of November 2026. No source gives a reason. ## Sources 1. Twill pricing page, read 12 November 2026 — https://twill.ai/pricing (grade C). Read in full. Source for the plans, the credit definition and the per-workspace pricing. 2. Twill home page, read 12 November 2026 — https://twill.ai/ (grade C). Read in full. Source for the product description, the integrations, the Y Combinator badge and the structured page data. 3. Launch HN: Twill.ai (YC S25) – Delegate to cloud agents, get back PRs, 10 April 2026 — https://news.ycombinator.com/item?id=47720418 (grade C). Read the post and the founders' replies. Source for the founders' first names, the architecture, the batch and the launch date. 4. Twill Terms of Service, effective 12 March 2026 — https://twill.ai/terms (grade C). Read in full. Source for the operator Anybase Inc. and the paid plans clause. 5. Twill Privacy Policy, effective 12 March 2026 — https://twill.ai/privacy (grade C). Read in full. Source for the third-party services, the Stripe customer ID and the operator. 6. Twill newsletter, 16 March 2026: CLI, BYOK, and a new way to score repos — https://twill.ai/newsletter/2026-03-16 (grade C). Read in full. Source for the move to token credit billing and the exemption for bring-your-own-keys. 7. Internet Archive copy of the Twill home page, 9 September 2025 — https://web.archive.org/web/20250909121727/http://twill.ai/ (grade C). Read in full. Source for the waitlist and the earlier plans. 8. Internet Archive copy of the Twill terms, 6 October 2025 (effective 15 August 2025) — https://web.archive.org/web/20251006173938/http://twill.ai/terms (grade C). Read in full. Source for the earlier operator Chainlit SAS. 9. Startup Fortune, 11 April 2026: Twill.ai Runs Coding Agents in the Cloud and Delivers PRs While You Sleep — https://startupfortune.com/twillai-runs-coding-agents-in-the-cloud-and-delivers-prs-while-you-sleep/ (grade C). Read the opening. It retells the launch post and names no customer. 10. Twill newsletter, 24 May 2026: Your AI engineers have persistent machines — https://twill.ai/newsletter/2026-05-24 (grade C). Read in full. Shows that the product was still changing in May 2026. 11. TwillAI agentbox-sdk repository on GitHub, read 12 November 2026 — https://github.com/TwillAI/agentbox-sdk (grade C). Read the repository page. Source for the licence and the commit count. 12. Database of failed startups, local export — /method/#dead (grade C). Record pages read on the website on 12 November 2026. AI-written summaries, not independently verified. Source for the failed predecessors. 13. Recherche d'entreprises, the French government's company register search: Chainlit (SIREN 951842749), read 12 November 2026 — https://recherche-entreprises.api.gouv.fr/search?q=chainlit (grade B). Read the one result in full. Source for the creation date of Chainlit SAS, its seat in Paris and its active status. 14. Chainlit repository on GitHub, read 12 November 2026 — https://github.com/Chainlit/chainlit (grade C). Read the notice at the top of the README. The creation date of the repository comes from GitHub's public data for it. 15. Internet Archive copy of the Literal AI home page, 3 September 2024 — https://web.archive.org/web/20240903024215/https://literalai.com/ (grade C). Read the opening. Source for the sentence that the team behind Chainlit built Literal AI. Contact: aiolum@livyden.com. Run Twill.ai? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: Twill.ai).