Back to vendors
E

Ellipsis

Also known as: ellipsis-dev, Ellipsis Agent Cloud

Visit site
Entry priceUsage based, with model tokens plus a 10% platform fee and no seat fees.Full pricing detail

Managed agent platform for coding agents defined as YAML in your repo, running Claude Code, Codex or Gemini CLI in isolated per session sandboxes with scoped credentials and hard spend caps, triggered by GitHub, Slack, Linear and Sentry events or cron, with every session recorded.

Ellipsis is a managed agent platform for coding agents, built by Ellipsis AI Inc. in New York and backed by Y Combinator. Its homepage pitches it for developers, software factories, code review and background agents, on any model and any harness. Teams define each agent as a YAML file in a repository. The file declares its trigger, the harness it runs (Anthropic's Claude Code, OpenAI's Codex or Google's Gemini CLI), its model, prompt, environment, permissions and budget. The version on the default branch is live, branches preview it, and a merge deploys it.

Agents start on cron schedules or on GitHub pull request, push, issue, check and release events, Linear issues, Slack channel creation and Sentry events, the last still short of general release. They can also start from an @ellipsis mention, the dashboard, the CLI or the REST API. Each session runs in its own isolated cloud sandbox with the repositories cloned in, up to 32 live sessions per account by default, and opens pull requests, posts reviews or answers in the thread that started it. Code review is one agent among them, set up as a pull request trigger, a prompt saying what to look for and an output setting that comments, approves or requests changes.

Ellipsis supplies the platform around the agent. Each session's GitHub credential is minted at the permissions its agent declares and revoked at teardown, and secrets are write only. Hard spend limits apply per session, agent, developer and account. Every session is recorded, with messages and tool calls visible live, transcripts downloadable and searchable, and cost tracked per session. Everything is scriptable through a REST API with Python and TypeScript SDKs, a CLI and Ellipsis's own MCP server.

Ellipsis runs as a managed service or inside the customer's own AWS VPC. It is priced on usage, with model tokens plus a 10% platform fee and sandbox compute at cost, and there are no per seat fees. It holds SOC 2 Type I, with Type II in progress.

From our cloud coding agents video

Devin, Factory, Jules, OpenHands: What Each Cloud Coding Agent Does Best

Our video on the 11 cloud coding agents in the index names what each one does best. For Ellipsis, that is agents as code, each one a file in your repository.

The player starts at Ellipsis's part, 7:37 in. Recorded in October 2026; the profile here is live and updates as research is verified.

Vendor details

Canonical URL

https://www.ellipsis.dev

Category

Coding agent

Subcategory

Cloud platform for coding agents, with AI code review

Funding status

Independent. Founded in 2024 by Nicholas Moy and James Murdza and backed by Y Combinator in the W24 batch, Ellipsis raised roughly two million dollars in seed funding. It reports use across more than twenty nine thousand codebases and by more than four hundred engineering teams, with named users including PromptLayer, Warp, and Onfleet.

Company status

independent

Use cases & customers

Primary use cases

automated pull request reviewagentic bug fixing and code generationcodebase question answeringdeploying and managing coding agents

Target customers

software engineering teams on GitHubstartups and small to medium engineering teamsopen source maintainers

Deployment options

SaaS (GitHub app + Agent Cloud)Per-session ephemeral Linux sandboxes, destroyed at completionPrivate instance in customer AWS, GCP or Azure (enterprise, by arrangement)

Integrations

A GitHub app with per session scoped tokens, pull request, push, issue, check and release triggers and @ellipsis mentions, plus Slack and Linear integrations, a Sentry integration not yet generally released and Jira listed as coming. Each agent's environment attaches MCP servers and secrets. A REST API with a published OpenAPI specification, Python and TypeScript SDKs, the agent CLI and an MCP server at mcp.ellipsis.dev start, stream, message and stop sessions and manage agents, environments, secrets and webhooks, and signed outbound webhooks report session events. Agents act by opening pull requests, posting reviews and answering in the thread or ticket that started them.

In practice

Pull request review has become the bottleneck. A code review agent on Ellipsis reviews every push to a non draft pull request and posts one review with comments on the lines it flags.

You want routine engineering chores handled on a schedule. Define an agent in one YAML file with a cron trigger, a model and a session budget, merge it, and it opens a pull request on each run while you watch its session record.

Your security team will not let agent code or LLM calls leave your cloud. The Your AWS Account offering runs the Ellipsis control plane and sandboxes in your own AWS VPC with the same agents and configs.

Agentic Index coverage score

10.5 / 14 capabilities · 75%

Integrations & Tool Calling Full

The GitHub App, Slack and Linear integrations let agents open pull requests, post reviews and answer in the thread or ticket that started them, and a Sentry integration is documented but not generally released, with Jira listed as coming. Each agent's environment attaches MCP servers (Slack, Linear, remote or local) and secrets for further tools, so the customer extends the tool surface. GitLab no longer appears on the current site.

Sourceellipsis.dev/docs/integrations and docs/environmentsread 2026-10-28

Workflow Orchestration Full

Agents are YAML files in the repository declaring trigger, harness, model, prompt, environment, permissions and budget, live on merge and previewed on branches. Parallel sessions each run in their own sandbox, branch and pull request, with a default of 32 live sessions per account, agents can start further sessions of their own, and a person can steer a session mid task. The documented harnesses are Claude Code, Codex and Gemini CLI; OpenCode, Copilot, Cursor, Pi, Grok Code and Antigravity are listed as not yet available, and no sub agent mechanism is documented.

Sourceellipsis.dev/docs/agents, docs/harnesses and ellipsis.dev/pricingread 2026-10-28

Knowledge Grounding & RAG Partial

Context is assembled per session. Each session checks out the repositories its environment names from a cached environment image, so the Claude Code, Codex or Gemini CLI harness reads the working tree directly, alongside Codex skills and connected Slack, Linear and Sentry context. There is no persistent index, graph or embeddings layer over the customer's code.

Sourceellipsis.dev/docs/environments and docs/sessionsread 2026-10-28

Human Oversight & Guardrails Partial

Ellipsis constrains what an agent can do rather than gating its actions. Permissions declared per agent limit each session's GitHub credential to named repositories and to read only or specific write scopes and set its Ellipsis API scopes, and hard budgets stop model requests per session, agent, developer and account before compute starts.

A code review agent's output is set to stay silent, comment or approve on a pass and to comment or request changes on a fail, with blocking comments marked, and work reaches the codebase as pull requests the customer merges on its own code host. There is no step where a person approves an agent action before it commits, and no plan review or auto merge.

Sourceellipsis.dev/docs/permissions, docs/budgets and docs/code-reviewread 2026-10-28

Security, Identity & Governance Full

SOC 2 Type I certification is in place, with Type II in progress and no trust page. Access controls cover owner and administrator roles for account settings with per developer spend limits, scoped API keys, per session GitHub credentials minted at the permissions each agent declares and revoked within an hour at teardown, write only secrets, zero source retention, and customer set session data retention from zero hours to one year with deletion within four hours of expiry. There is no SSO, SAML or SCIM support.

Sourceellipsis.dev/docs/security, platform/security and docs/api/authenticationread 2026-10-28

Observability & Auditability Full

Every session is recorded. The dashboard and CLI show messages and tool calls as they happen, session events, turns and records are readable over the API, and an audit trail covers commands, file changes and tool calls. Transcripts are downloadable as unmodified Claude Code, Codex and Gemini CLI output, sessions are searchable and linked to the pull requests they produced, code review history is kept in the dashboard, and cost and token, CPU and memory usage are tracked per session, developer and agent.

Sourceellipsis.dev/docs/sessions and platform/observabilityread 2026-10-28

Memory & State Persistence Partial

A session stays open for follow ups that continue the same workspace, and a mention starts a durable conversation per pull request, issue or Slack thread that later replies continue, with session records kept for a customer set retention of up to a year. That state stays within one thread. No memory layer with its own scope and lifetime that the agent reads across independent sessions is documented.

Sourceellipsis.dev/docs/sessions and docs/agents/triggersread 2026-10-28

Deployment & Data Residency Full

Ellipsis comes in two offerings. With Managed SaaS, each session runs in its own sandbox VM in Ellipsis's cloud, sized by the agent's compute setting, and Your AWS Account deploys the control plane and agent sandboxes into the customer's AWS VPC with no credentials, code or LLM calls leaving it, billed by AWS. Session data retention is set by the customer. No region choice is stated.

Sourceellipsis.dev/pricing and docs/environmentsread 2026-10-28

Prebuilt Agents, Templates & Packs Partial

A code review agent is available as example YAML with documented output settings, Codex sessions take up to ten skills (Claude Code and Gemini skills are not supported), and skills install with npx skills add. The docs describe no template gallery, prebuilt agent library or review rule packs, only use case pages and guides.

Sourceellipsis.dev/docs/code-review and docs/agentsread 2026-10-28

Triggers & Channel Coverage Full

Each agent carries one trigger: a cron schedule in UTC, or an event on GitHub pull requests (opened, pushed, merged, closed, review submitted, commented), pushes, issues (opened, labeled), check runs, check suites, workflow runs, statuses and releases, Linear issues, Slack channel creation or Sentry events, and each run starts a session with no one invoking it. Sessions also start on demand from @ellipsis mentions, the dashboard, the CLI and the API. Issue assignment is not a trigger, and webhooks are outbound only.

Sourceellipsis.dev/docs/agents/triggersread 2026-10-28

Model Flexibility & Routing Full

The customer picks the harness and model for every agent and session across three makers: Claude models in Claude Code, GPT models in Codex and Gemini models in the Gemini CLI, each priced per million tokens. A Claude or ChatGPT subscription can pay for sessions a user starts. No bring your own key path is documented.

Sourceellipsis.dev/docs/models and docs/pricingread 2026-10-28

APIs, SDKs & MCP Extensibility Full

A REST API at api.ellipsis.dev/v1 with bearer API keys and a published OpenAPI specification starts, streams, messages and stops sessions and manages agents, environments, secrets and webhooks. Python and TypeScript SDKs and the ellipsis CLI work over the same surface.

Sourceellipsis.dev/docs/api and docs/api/authenticationread 2026-09-29

Testing, Debugging & Optimization Partial

Ellipsis tests the customer's code, not the agent. A code review agent reviews each push to a pull request and runs the project's test suite in its sandbox, an environment validate hook and an Ellipsis environment pull request check confirm the setup, and session records show the tests run. There is no harness, scored test set or optimization loop for evaluating agent behavior.

Sourceellipsis.dev/docs/code-review, docs/environments and docs/sessions; ellipsis.dev/docs/agents/code-reviewread 2026-10-28

Browser & Computer Use Not documented

Ellipsis does not operate browsers or desktops. Sessions run the Claude Code, Codex or Gemini CLI harness in an isolated cloud Linux sandbox with terminal and file access, which is code execution, no browser, desktop or computer control tool is offered, and Gemini's web search and fetch are turned off.

Sourceellipsis.dev/docs/environments and docs/harnessesread 2026-10-28

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Usage based, with model tokens plus a 10% platform fee and no seat fees.

Usage based. Prepaid credits pay model tokens at the published rate per million tokens, with a 10% platform fee on credit purchases, and organization sandbox CPU and memory at cost. Personal compute is free, and there are no per seat fees.

Free tier

Included quota

No included quota. Every session is paid from prepaid credits or, for sessions a user starts, the user's own Claude Code or Codex subscription. Both offerings include the full platform with no feature gates or tiers.

What is public

Ellipsis publishes its whole pricing model. There are no per seat or idle fees, and agents cost model tokens plus a 10% platform fee, with token prices listed per model per million tokens (for example Claude Opus 5.5 at $4 in and $20 out, Sonnet 5.5 at $2 and $10, GPT-6.1 Sol at $2 and $10, Gemini 4 Argon at $2 and $10). Organization sandbox compute is $0.142 per vCPU-hour and $0.024 per GB-hour, charged while a sandbox waits, and personal compute is free. A Claude or ChatGPT subscription can pay for sessions a user starts. Two offerings run the full platform, Managed SaaS and Your AWS Account. In Your AWS Account, tokens and compute are billed by AWS and the 10% platform fee applies to token cost. Optional support packages are Standard $5,000, Advanced $10,000 and Premier $15,000 a month, and startups are offered special pricing by email.

Billing mechanics

Credits are bought in the dashboard, and a 10% platform fee applies to credit purchases. Model tokens are charged at each model's published rate, and organization sandbox CPU and memory at cost while a sandbox runs, including while it waits for a message. Scheduled and event triggered agents and API key sessions always use credits. Automatic recharge is optional and credits do not expire.

Cost watchouts

Idle sandboxes accrue CPU and memory charges while waiting for a message. The 10% platform fee is charged on credit purchases. Large context GPT-6 requests over 272K input tokens cost double input and one and a half times output.

Variable cost rationale

Every session is metered on model tokens, at each model's published rate plus a 10% platform fee, and on organization sandbox CPU and memory. There is no flat plan or seat, so spend grows with model choice, session length, sandbox size and how often agents run, and a sandbox waiting for a message still accrues compute. Spend limits per session, agent, developer and account cap the bill, and paid requests stop when the prepaid balance or a limit runs out. Personal compute is free, and sessions a user starts can run on that user's own Claude Code or Codex subscription.

Additional watchouts

Cost scales with model choice, session length and sandbox size, and a sandbox waiting for the next message still accrues compute. Scheduled and event triggered agents cannot use a personal subscription. Spend limits per session, agent, developer and account are the control.

Overage / add-ons

Paid requests stop when the credit balance or an applicable spend limit is exhausted. Automatic recharge is optional, and credits do not expire.

Sales call required

No, self serve available

Free / trial

Personal compute is free, and sessions a user starts can run on their own Claude Code or Codex subscription. Other usage is paid from credits.

Lowest paid plan

No plans or seats. Prepaid credits pay model tokens plus a 10% platform fee, and organization compute at $0.142 per vCPU-hour and $0.024 per GB-hour.

Commercial notes

Fully self serve with no seats. Enterprises can add a support package for a Slack channel, onboarding, help with deployment in their own AWS account and security questionnaires.

Key ambiguities

Real cost depends on model, session length and sandbox size rather than on seats. The two pricing pages state the 10% fee on different bases: the pricing page puts it on token cost, the docs put it on credit purchases before discounts.

Cancellation / refund

Prepaid credits do not expire, and there are no seats or contract. No refund terms are published.

Support SLA / resale

Support packages at $5,000, $10,000 and $15,000 a month add a Slack support channel, onboarding, BYOC deployment help, security questionnaires and first response times of 48, 24 and 4 hours. Ellipsis is SOC 2 Type I certified, with Type II in progress.

Missing data

Every rate is published. Enterprise pricing beyond the support packages is not described.

Agentic Index verified 2026-10-28

Alternatives to Ellipsis

The closest documented capability profiles to Ellipsis among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.

  • Poolside11.0 / 14Fuller documented coverage on Human Oversight & Guardrails
  • OpenCode10.5 / 14Fuller documented coverage on Human Oversight & Guardrails
  • Tabnine11.5 / 14Fuller documented coverage on Knowledge Grounding & RAG and Human Oversight & Guardrails
  • Augment Code12.0 / 14Fuller documented coverage on Knowledge Grounding & RAG and Human Oversight & Guardrails
  • Charm10.0 / 14Fuller documented coverage on Human Oversight & Guardrails
  • CodeRabbit12.0 / 14Fuller documented coverage on Knowledge Grounding & RAG and Human Oversight & GuardrailsEllipsis vs CodeRabbit →

Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.