cubic
Also known as: cubic, cubic.dev, cubic-dev-ai, mrge
AI code review agent that reviews every pull request with whole-repository and cross-repo context, learns from reviewer corrections, runs scheduled codebase scans that file tickets and open fix PRs, and can auto-approve clean PRs under a policy you test in shadow mode first.
cubic is an AI code review platform built for the moment when AI-generated code made review the bottleneck. When a pull request opens, cubic reviews the changes automatically in a secure isolated container with context into the whole codebase rather than just the diff, and comments inline on the lines that matter.
Its reviewer navigates a repository the way a developer would, and its analysis is repository-wide, so it catches issues that span multiple files. Cross-repo reviews extend that further, letting cubic read selected companion repositories when a change to a frontend affects a backend API or a shared schema.
A Y Combinator backed startup, listed there under its earlier name mrge, cubic is used by teams at n8n, Cal.com, PostHog, Granola, Resend, Daytona and the Linux Foundation.
Beyond flagging bugs, cubic learns each team's conventions: reacting to a suggestion or correcting it is remembered, which reduces false positives over time. Teams define custom agents in natural language or code patterns to enforce internal API usage, architectural decisions or deprecated functions, and cubic has access to current documentation for the libraries in the stack, so suggestions reflect present APIs rather than training-cutoff patterns.
It generates pull request descriptions, and offers fixes: click Fix with cubic or tag the agent in a comment and background agents generate and apply them, with heavier work handed to a coding agent such as Claude Code or Cursor. Ultrareview escalates risky migrations and security-sensitive changes to the most capable review models on request. On higher tiers, scheduled codebase scans comb whole repositories for bugs and security issues, triage findings, notify owners over Slack and email, create tickets in Jira, Linear, Asana or Notion, and open one-click fix pull requests.
Two capabilities set the oversight story apart. Auto-approval lets cubic approve clean pull requests under a policy the repository owner defines, and ships with a shadow mode that shows which pull requests cubic would have approved before live mode is switched on, so a team validates the policy before granting the agent any authority.
And cubic runs outside GitHub as well as inside it: a CLI installed by shell script reviews uncommitted changes, branches or specific commits before a pull request exists, and cubic connects to Cursor, Claude Code, VS Code, Codex and Gemini CLI to run local reviews and answer codebase questions without leaving the editor. An AI wiki indexes the repository into searchable documentation with architecture diagrams on demand, refreshed daily or weekly.
cubic exposes its own work to other systems through two MCP servers, one for codebase scan findings and one for the AI wiki, plus an Analytics API. It tracks model attribution per commit, so it can report which coding agent wrote a change and whether bugs were found in it.
cubic is SOC 2 compliant by its own statement, reviews code without storing it or using it to train models, and adds SSO and SAML through GitHub OAuth, GitHub Enterprise support, exportable compliance audits, bring-your-own API keys and custom MSA and DPA terms at the Enterprise tier. No trust center or published attestation report was found, so the certification rests on vendor assertion.
Pricing is per developer. The free Starter plan includes 20 pull request reviews a month, up to five custom agents, automatic descriptions, project connectors and unlimited AI wikis. Team is $30 a month billed annually or $40 monthly, adding 40,000 reviewed lines per developer, the CLI, Ultrareview and simple analytics. Pro is $79 annually or $99 monthly, adding 80,000 reviewed lines, nightly scans on three repositories, auto-created fix pull requests, Confluence, the MCP servers and the Analytics API. Enterprise is custom. cubic is free for public repositories.
Vendor details
Canonical URL
https://www.cubic.dev
Category
Coding agent
Subcategory
AI code review agent
Funding status
Independent. Y Combinator backed, listed under the YC company name mrge, which still appears in the pricing page's backing link and enterprise contact metadata. Customers named on first-party pages include n8n, Cal.com, PostHog, Granola, Resend, Daytona, Legora, Better Auth, Cartography and the Linux Foundation. Specific funding figures are not disclosed on first-party pages.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Two-way GitHub integration installs as an app in two clicks and drives inline comments, fix branches and auto-approval, with GitHub Enterprise support at the Enterprise tier. Project connectors cover Jira, Linear, Asana and Notion from the free plan, with Confluence added on Pro, so a change referencing a ticket is reviewed with that context. Slack and email notifications ship on Pro as outputs. cubic also runs inside Cursor, Claude Code, VS Code, Codex and Gemini CLI, ships a local CLI installed by shell script, and exposes two MCP servers (Codebase scan MCP and AI wiki MCP) plus an Analytics API so other agents and systems can consume its findings. The agent acts by commenting inline, committing one-click fixes, opening fix pull requests and creating tickets.
In practice
Your team is drowning in pull requests and rubber stamping them. cubic reviews each PR against your whole codebase, flags real bugs on the exact lines, and enforces the rules you wrote in plain English.
You maintain a busy open source project alone. cubic is free for public repos, learns your conventions, and reviews contributor PRs so you spend less time on routine feedback.
You worry about latent bugs already merged. On higher tiers cubic runs nightly scans where many agents comb the repo for bugs and vulnerabilities, then file tickets and open fix pull requests.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
9.5 / 14 capabilities · 68%
| Integrations & Tool Calling | Full |
|---|---|
|
A two way GitHub app drives inline comments, fix branches and approvals, project connectors cover Jira, Linear and Asana with Confluence on Pro, Slack support ships on Pro, and cubic runs inside coding agent hosts through its MCP server and CLI. Sourcecubic.dev/pricing and pricing-plans; cubic.dev/pricing-plansread 2026-09-29 |
|
| Workflow Orchestration | Partial |
|
The review runs a fixed sequence (review, inline comments, fix branches and optional approval), with scheduled codebase scans that triage and file tickets, and implementation can be handed to an outside coding agent. No branching, conditions or coordination across cubic's own agents is documented. Sourcedocs.cubic.dev llms.txt (codebase scans, coding agents) and cubic.dev/pricing-plansread 2026-09-29 |
|
| Knowledge Grounding & RAG | Full |
|
The AI wiki automatically indexes the codebase into searchable documentation with source links and architecture diagrams, regenerated weekly or monthly as code changes and queryable in plain English, over MCP or as synced markdown, alongside whole repository and cross repository context during reviews. Sourcedocs.cubic.dev/wiki/ai-wikiread 2026-09-29 |
|
| Human Oversight & Guardrails | Partial |
|
Auto approval runs under a per repository policy the customer sets (low risk only, custom or always), and cubic approves only when its review finds no issues. The docs state no person approves anything before cubic acts, so the policy is a customer constraint on the agent rather than an approval step. Shadow mode, which previews approvals before they go live, is covered under testing. Sourcedocs.cubic.dev/ai-review/auto-approvalread 2026-09-29 |
|
| Security, Identity & Governance | Full |
|
The privacy and security page states SOC 2 Type 1 compliance with TLS 1.2 or higher in transit and AES-256 at rest and no training on code, and Enterprise lists SSO and SAML through GitHub OAuth, GitHub Enterprise support, exportable compliance audits and custom MSA and DPA terms. Sourcedocs.cubic.dev/account/privacy-security and cubic.dev/pricing-plansread 2026-09-29 |
|
| Observability & Auditability | Partial |
|
The Analytics API returns per pull request metrics (issues flagged and fixed, AI authored lines, engagement times), Pro tracks AI coding usage, and Enterprise exports compliance audits; these report outcomes, and no per run record of the review agent's steps or tool calls is documented. Sourcedocs.cubic.dev/analytics/api and cubic.dev/pricing-plansread 2026-09-29 |
|
| Memory & State Persistence | Full |
|
cubic extracts learnings from PR comments, reactions and senior reviewers' past reviews, scoped to the team and never shared across organizations or repositories. Learnings are browsable, editable and deletable under Memory and Learning, and a weekly cleanup merges duplicates and archives outdated guidance. So the memory has a stated scope, a lifetime, a review path and a deletion path. Sourcedocs.cubic.dev/ai-review/memory-and-learningread 2026-09-29 |
|
| Deployment & Data Residency | Not documented |
|
cubic is hosted SaaS installed as a GitHub app, with code stated never to be stored or used for training; no self hosting, customer cloud or region choice is documented on the pricing, enterprise or privacy pages. Sourcecubic.dev/pricing-plans, cubic.dev/enterprise and docs privacy-securityread 2026-09-29 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
Custom review agents are written by the team in natural language to enforce its own standards, five on Team and ten on Pro, so the assets are ones the customer assembles; no library of prebuilt agents or templates from cubic is documented. Sourcecubic.dev/pricing and docs.cubic.dev llms.txt (custom agents)read 2026-09-29 |
|
| Triggers & Channel Coverage | Full |
|
Reviews run automatically when a pull request opens and again on new commits, codebase scans run nightly on Pro, and the AI wiki regenerates on a weekly or monthly schedule, so work reaches the agent without a person starting it. @cubic mentions, Fix with cubic and the CLI add requested runs. Sourcecubic.dev/pricing-plans and docs.cubic.dev/wiki/ai-wikiread 2026-09-29 |
|
| Model Flexibility & Routing | Partial |
|
The privacy page names OpenAI and Anthropic as the model providers cubic runs on, so the vendor discloses more than one provider but keeps the choice itself. Enterprise lists bring your own API keys, but no model or provider selection comes with them, and supplying a key is not choosing a model. Sourcedocs.cubic.dev/account/privacy-security and cubic.dev/pricing-plansread 2026-09-29 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
cubic's own MCP server at cubic.dev/api/mcp, signed in with OAuth or an API key, lists 22 tools including get_pr_issues, trigger_pr_review, update_pr_issue_status, list_wikis, list_scans and get_scan. Several of them start reviews or change finding status, so outside agents can act through cubic, not only read from it. A Pro Analytics API adds a documented bearer token endpoint. Sourcedocs.cubic.dev/ide/mcp-server and analytics/apiread 2026-09-29 |
|
| Testing, Debugging & Optimization | Full |
|
Shadow mode keeps cubic in comment only mode on real pull requests while adding a summary of whether it would have approved each one, so a team tests the agent's approval behavior under its chosen policy before granting it live authority. Bug detection, the product itself, tests the customer's code rather than the agent. Sourcedocs.cubic.dev/ai-review/auto-approvalread 2026-09-29 |
|
| Browser & Computer Use | Not documented |
|
cubic works through the GitHub API, an isolated review container, a CLI and MCP; no browser control or operation of software through a graphical interface is documented. Sourcedocs.cubic.dev and cubic.dev/pricing-plansread 2026-09-29 |
|
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
Recent platform changes
cubic added a repository policy that allows automatic pull request approval while findings at selected priorities remain unresolved. Choosing a priority also allows lower priorities, and cubic publishes new findings before submitting a separate approval. Other approval rules and GitHub branch protections still apply.
Bears on: Human approval / guardrails
View sourcecubic now supports cross-repo reviews, allowing users to link up to five related repositories for coordinated updates. The agent checks related code across repository boundaries, displays findings inline with citations to affected files, and tracks fixes across companion pull requests.
Bears on: Agent capability
View sourcePricing
Free · Team $30/dev/mo billed annually ($40 monthly)
Per-developer/month subscription tiers with a monthly reviewed-line allowance (added/deleted diff lines cubic reads); free tier capped by monthly review count; enterprise custom
Included quota
Per-developer pricing with reviewed-line allowances. Starter free: 20 PR reviews/month, up to 5 custom agents, automatic PR descriptions, custom context, Jira/Linear/Asana/Notion connectors, unlimited AI wikis. Team $30/developer/month billed annually or $40 monthly: 40,000 reviewed lines per developer, cubic CLI, Ultrareview, weekly wiki updates, simple analytics. Pro $79 annually or $99 monthly: 80,000 reviewed lines per developer, faster PR and CLI reviews, up to 10 custom agents, Fix with coding agents, Confluence, nightly scans on 3 repositories, auto-created fix PRs, Slack and email notifications, Codebase scan MCP, AI wiki MCP, 20 private repo wikis, AI coding usage tracking and an Analytics API. Enterprise custom: additional custom agents and scans, SSO/SAML with GitHub OAuth, GitHub Enterprise support, exportable compliance audits, bring your own API keys, custom MSA, DPA and payment terms. Free and unlimited for public repositories.
What is public
All tier prices are published exactly on a per-developer basis, with annual and monthly rates shown side by side: Free $0, Team $30 annual or $40 monthly, Pro $79 annual or $99 monthly, Enterprise custom. Reviewed-line allowances (40k Team, 80k Pro), custom agent caps (5 and 10), scan cadence and per-tier feature lists are all published, as is the definition of a reviewed line. Only Enterprise pricing is quote-based.
Billing mechanics
Per-developer/month tiers with monthly reviewed-line allowances; free tier capped by review count; free & unlimited for public repos.
Cost watchouts
Reviewed-line allowances can be exceeded by high PR volume/large diffs; Pro roughly doubles+ the per-seat cost; free tier caps at 20 reviews/mo.
Variable cost rationale
Per-developer pricing is predictable, but each tier caps reviewed lines per month (e.g. ~40,000 on Team) and the free tier caps monthly review count, so high PR volume or large diffs can push a team to a higher tier. The main cost step is moving to Pro (~$79-99) for nightly scans and auto-fix. Exposure is moderate: bounded by per-seat tiers but sensitive to review volume and team size.
Additional watchouts
Nightly background scans (the standout Pro feature) were still maturing/early-access in early 2026. Free tier review cap (20/mo) hits quickly for active teams. Pro (~$79-99/dev) can be a stretch for small teams. GitHub-centric. Compliance-strict teams should verify data handling.
Overage / add-ons
Usage is metered by reviewed lines (added/deleted diff lines cubic reads) against the monthly allowance; the free tier is capped by monthly PR-review count. Excess beyond allowance requires a higher tier.
Sales call required
No, self serve available
Free / trial
Free Starter plan: 20 PR reviews/month, up to 5 custom agents, automatic PR descriptions, custom context, Jira/Linear/Asana/Notion connectors and unlimited AI wikis. Unlimited and free for public repositories. Paid tiers carry a 7-day trial.
Lowest paid plan
Team $30/developer/month billed annually upfront, or $40/developer/month monthly
Commercial notes
Priced above CodeRabbit; positioned on review depth (repo-wide context, custom rules, background scans). Free tier genuinely usable (esp. open source). Pro's nightly scans are the premium draw. Value highest where review quality is inconsistent or seniors are bottlenecked.
Key ambiguities
Reviewed lines are the metered unit and the vendor defines them precisely: the added and deleted lines in a diff that cubic actually reads, excluding generated files, vendored code, binaries, ignored and skipped files and failed or canceled reviews, with incremental reviews counting only new changes and manual reruns counting again. Usage resets with the billing period. Two Pro variants appear on the page with differing scan cadence and trial length, one listing weekly scans on 5 repositories and a 14-day trial and the other nightly scans on 3 repositories and a 7-day trial, so the exact Pro entitlement should be confirmed at purchase.
Cancellation / refund
Free tier and 2-week trial (no credit card). Monthly per-developer tiers. Enterprise via MSA/DPA. Specific cancellation/refund terms not detailed.
Support SLA / resale
Standard support on paid tiers; Enterprise adds custom MSA/DPA. SOC 2 compliant; code reviewed in a secure container, never stored or used for training. No reseller/whitelabel program surfaced.
Missing data
Per-reviewed-line economics beyond the allowances and Enterprise pricing are not public.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose specialized AI agents build, host,…
- AgentUI — Managed AI app builder for operations teams that generates and hosts…
- Aider — Open source, model agnostic terminal coding agent that edits your…
- Anthropic Claude Code — Anthropic's agentic coding system across terminal, desktop, IDE, web…
- AppFactor — Agentic platform that runs persistent agents across an enterprise…
Alternatives to cubic
The closest documented capability profiles to cubic among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Sonar10.5 / 14Adds documented Deployment & Data Residency
- Greptile11.0 / 14Adds documented Deployment & Data Residency
- Qodo12.0 / 14Adds documented Deployment & Data Residency
- Graphite7.5 / 14Fuller documented coverage on Workflow Orchestrationcubic vs Graphite →
- CodeRabbit11.0 / 14Adds documented Deployment & Data Residencycubic vs CodeRabbit →
- Blink9.5 / 14Adds documented Deployment & Data Residency and Browser & Computer Use
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded