Ellipsis (ellipsis.dev) is a YC-backed company that launched in 2023 as an automated PR review and bug-fix bot and, as of July 2026, has pivoted to the Ellipsis Agent Cloud, a managed cloud platform for running coding agents such as Claude Code and Codex in governed sandboxes.
What it is #
Ellipsis AI Inc (New York, founded 2023, YC Winter 2024, founders Hunter Brooks and Nick Bradford) sells a platform where every coding agent is a YAML file in your repo, and each session runs in its own sandbox with scoped credentials, hard budget caps, and searchable transcripts.
Sessions start from GitHub, Slack, Linear, or Sentry events, cron schedules, the agent CLI, or a REST API, and the platform fronts Claude Code, Codex, Copilot, Gemini, Cursor, Grok Code, Antigravity, OpenCode, and Pi.
Code review survives as a configurable agent use case, not the fixed install-and-forget bot of 2024, and the platform deploys either in Ellipsis’s cloud or into your own AWS VPC.
The core product is proprietary: its GitHub org publishes only tooling, an agent CLI, a homebrew tap, a Python and TypeScript SDK mirror, and the blink codebase-search repo, five public repos in total as of 2026-10-11, down from six on 2026-09-21 because the AWS installer repo is no longer public.
Blink (100 stars as of 2026-10-11) was created on September 16, 2026.
Status #
Active and pivoted.
The July 28, 2026 post “Introducing the Ellipsis Agent Cloud” says it plainly: Ellipsis “launched in 2023 as one of the first AI Code Review bots”, but teams no longer want a product that “just worked”, so the company rebuilt itself as agent infrastructure and released the agent CLI.
The homepage, docs, and pricing all describe the agent cloud as of 2026-10-03 (re-verified 2026-11-08, with the hero still leading Software Factories ahead of code review and background agents), and the GitHub org pushed to its CLI repo on 2026-11-07.
Funding is a $2M seed announced June 19, 2024, with YC partners and the Pioneer Fund among the investors; I found no later round on any page I fetched.
The blog is thin, three original posts since January 2025 and a gap from April 2025 to July 2026, so public cadence evidence is sparse even though the product is clearly alive.
The community footprint is one strong Show HN from May 9, 2024 (Algolia item 40309719, 121 points, 64 comments) and a quieter February 2024 one (item 39526616, 18 points, 11 comments).
Strengths #
- Agents as code: each agent is a versioned YAML file changed by pull request, so configuration gets reviewed like any other code.
- Governance is built in: per-session scoped tokens, budget caps enforced before a sandbox starts, and full searchable transcripts of every session.
- Harness-agnostic: the same interface fronts Claude Code, Codex, OpenCode, and others, so the platform layer does not lock you to one agent vendor.
- Security posture: zero source retention (code is deleted with the sandbox), SOC 2 Type I certified with Type II in progress as of 2026-09-18, and a BYOC path into your own AWS VPC.
- The founders were candid in 2024 that frontier LLMs were not good enough to write code autonomously, only to review it, which matched the product to what models could actually do.
Cautions #
- The 2024 review bot no longer exists as a product; the launch post says the install-and-forget model “is not the case anymore”, so getting review out of Ellipsis means writing and maintaining YAML.
- The public footprint is thin: a ten-month blog gap, five public repos, and an agent CLI at 3 stars as of 2026-10-02, so third-party validation is limited.
- Headline claims are company data: “merge pull requests 13% faster” comes from the seed post, and “most popular AI Code Review agent among YC founders” comes from the company’s own launch post.
- HN critics were blunt about the review era, calling the PR summaries noise and telling maintainers “please stop making open source worse”.
- The founder admitted on that thread that the company had no hard numbers (“all our data is qualitative”) as of May 2024.
- Cost risk sits with you: token spend is usage-based on top of metered compute, and optional support packages start at $5,000/month.
Pricing #
Tokens are billed at cost with no markup, plus a 10% platform fee, with CPU at $0.142/vCPU-hour and memory at $0.024/GB-hour, as of 2026-10-04 (re-verified unchanged 2026-10-25). There are no per-seat fees and no idle charges, and the $100 credit I recorded at launch in September 2026 is now displayed on no surface I can find: the homepage and pricing page omit it, and as of 2026-11-08 the GitHub org profile’s restored description (Cloud Agent Platform for Developers) and the YC directory’s rewritten agent-cloud pitch (a Managed Agent platform one-liner) both still omit it, so the credit is best read as gone. The free-for-individuals on-ramp is gone as of 2026-10-25: the homepage carries no free badge at all (the word free does not appear on the page), and the pricing FAQ now answers only a free-for-open-source question, case by case by email. Until 2026-10-17 the homepage led with a FREE FOR INDIVIDUALS badge for people on their own Claude Code or Codex subscription, and that wording had flip-flopped before it (“ChatGPT or Claude” on 2026-09-13, “Claude Code or Codex” by 2026-09-18, back on 2026-09-27, then just FREE FOR INDIVIDUALS by 2026-10-02), with the pricing FAQ naming Claude Code or Codex subscriptions for interactive sessions throughout. The same published pricing covers the managed SaaS and a BYOC deployment into your own AWS account. Optional enterprise support packages run $5,000 (Standard), $10,000 (Advanced), and $15,000 (Premier) per month as of 2026-10-04. This replaces the May 2024 model, which was $20/seat/month for the review bot.
Price history #
| Date | Plan | Change | Source |
|---|---|---|---|
| 2024-05 | Review bot | Original model: $20/seat/month for the review bot. | |
| 2026-09-18 | All tiers | Replaced with usage billing: tokens at cost plus a 10% platform fee, CPU $0.142/vCPU-hour, memory $0.024/GB-hour, $100 starting credit, no seat fees; enterprise support packages at $5,000/$10,000/$15,000 per month. | |
| 2026-10 | All tiers | Platform fee raised from 10% to 20% of token cost; CPU, memory, and support-package prices unchanged. | |
| 2026-10-03 | All tiers | Platform fee returned to 10% of token cost, reversing the 20% rate recorded the previous day; CPU, memory, and support-package prices unchanged. | |
| 2026-10-25 | Free individuals on-ramp | Removed from the homepage and the pricing FAQ; the only advertised free path is case-by-case open-source support by email. |
Compared to #
Greptile is the closest sibling that stayed a review product: it reviews PRs against the whole codebase without you owning agent config, which is exactly what Ellipsis gave up. CodeRabbit and Qodo sell finished review and test agents, so try them first if you want zero YAML. Ellipsis’s 2026 competition is not review bots but agent infrastructure: the company itself publishes compare pages against running Claude Code on GitHub Actions, which is the option many teams will try first because their organization already pays for it.
Bottom line #
Recommended for teams running fleets of coding agents that need budgets, permissions, and searchable logs around Claude Code or Codex. Not for teams that want an install-and-forget AI reviewer, because Ellipsis no longer sells that product. I will take the unpopular position that leaving review was the right call: a fixed review bot was always a feature, not a company, and Ellipsis read that earlier than most of the 2024 class.
Changes #
- 2026-08-30 - Created in the new Code review category with status set to pivoted after the July 2026 Agent Cloud launch.
- 2026-09-02 - Added the public SDK mirrors to the recorded GitHub org surface.
- 2026-09-09 - Style sweep split a two-sentence Bottom-line line.
- 2026-09-12 - Advanced support package raised from $7,500 to $10,000 per month.
- 2026-09-13 - Added the free-for-individuals on-ramp (own ChatGPT or Claude subscription) to the pricing picture.
- 2026-09-18 - The free-individual on-ramp now names Claude Code or Codex subscriptions (was ChatGPT or Claude), and the GitHub org grew to six public repos with the new blink codebase-search repo.
- 2026-09-20 - Added the Price history section tracking price changes in a table, per the new owner rule.
- 2026-09-26 - The GitHub org dropped to five public repos (the AWS installer repo is no longer public) and blink grew from 31 to 81 stars; pricing re-verified unchanged.
- 2026-09-27 - The homepage’s free-tier badge wording flipped back to ChatGPT or Claude (the FAQ keeps Claude Code or Codex) and blink grew from 81 to 84 stars; prices re-verified unchanged.
- 2026-10-02 - The platform fee doubled from 10% to 20% of token cost (CPU, memory, and support packages unchanged), with a price-history row added; the homepage badge now reads FREE FOR INDIVIDUALS with the FAQ keeping Claude Code or Codex, blink grew to 92 stars, and a banned-term word was reworded out of the Compared-to prose.
- 2026-10-02 - Repaired the References section, which a previous run had left truncated mid-line, completing the Show HN citation, and re-fetched every cited URL, all 200 this run.
- 2026-10-03 - The platform fee returned to 10% of token cost, reversing the 20% rate recorded the previous day, with a price-history row added, and blink grew from 92 to 93 stars while the CLI repo was pushed again on 2026-10-02.
- 2026-10-07 - The $100 starting credit no longer appears on the homepage, pricing page, or docs pricing page (it last showed there on 2026-10-04), so the pricing sentence now dates the claim; every other price re-verified unchanged, blink at 99 stars.
- 2026-10-17 - Rotation re-verification: every cited URL re-fetched live, all prices unchanged; the $100 credit is now claimed on the GitHub org profile and the YC directory description while the product’s own pages still omit it, and the CLI repo’s latest push moved to October 13.
- 2026-10-25 - Rotation re-verification: recorded the free-for-individuals on-ramp leaving the product’s own pages (no free badge anywhere on the homepage, the pricing FAQ now offering only case-by-case open-source support by email) with a price-history row, the hero now leading Software Factories, and the CLI repo’s push moved to October 23; every cited URL re-fetched live, all metered prices unchanged.
- 2026-11-01 - Rotation re-verification: every cited URL re-fetched live; all metered prices unchanged (tokens at cost plus a 10% platform fee, CPU $0.142/vCPU-hour, memory $0.024/GB-hour, support packages $5,000, $10,000, and $15,000), the org still holds five public repos with the CLI pushed October 31, and the blog’s newest post is still July 28, 2026; the $100 starting credit is now displayed on no fetched surface (the GitHub org profile’s description field is empty and the YC directory omits it), recorded in Pricing.
- 2026-11-08 - Rotation re-verification: every cited URL re-fetched live; all metered prices unchanged (the 10% platform fee, the compute rates, and the three support packages re-read from the pricing page), the org still holds five public repos with the CLI repo pushed November 7 and the SDK mirror November 8, the org profile’s description field is restored (Cloud Agent Platform for Developers) and the YC directory is rewritten to the agent-cloud pitch (a Managed Agent platform one-liner, spawn, manage, and govern cloud agents for Software Factories, custom Code Review, and Background Tasks), the $100 credit still appears on no fetched surface, the blog’s newest post is still July 28, 2026, and the Show HN thread is unchanged (121 points, 64 comments); recorded in Status and Pricing.
See also #
- Greptile - the AI code review sibling that stayed a review product, the closest match to what Ellipsis abandoned.
- OpenCodeReview - another AI review entry in this section, useful to contrast productized review with agent-defined review.
- Agentic Coding Tools Landscape - the tracker that situates agent clouds like Ellipsis in the wider tooling map.
- Software Factory Feature Matrix - the feature matrix for this category, where agent cloud platforms are compared row by row.
References #
https://www.ellipsis.dev/ - current product surface: agent cloud framing, SOC 2 status, supported agents, fetched 2026-10-02, re-fetched 2026-10-25, 2026-11-01, and 2026-11-08 (no $100 credit and no free-for-individuals badge anywhere on the page; the hero still leads Software Factories).
https://github.com/ellipsis-dev - the public GitHub org: five repos (TalkFormAI, homebrew-cli, the agent CLI at 3 stars pushed 2026-11-07, the SDK mirror pushed 2026-11-08, and blink at 100 stars as of 2026-10-11), fetched 2026-10-17, re-fetched 2026-10-25, 2026-11-01, and 2026-11-08 (the org profile’s description was empty on 2026-11-01 and is restored by 2026-11-08, reading Cloud Agent Platform for Developers, still with no $100 credit line).
https://www.ycombinator.com/companies/ellipsis - YC batch (W24), founding date, founders, team size, and active status, fetched 2026-09-18, re-fetched 2026-10-25; on 2026-11-01 the directory description no longer named the $100 credit, and by 2026-11-08 it is rewritten to the agent-cloud pitch (a Managed Agent platform one-liner and a long description naming Software Factories, custom Code Review, and Background Tasks), still without it.
https://hn.algolia.com/api/v1/items/40309719 - the May 9, 2024 Show HN (121 points, 64 comments) with the praise and the blunt criticism that greeted the launch, re-fetched 2026-10-25 and unchanged 2026-11-01 and 2026-11-08.