Plan before building
Composer reads the code and presents a proposed change plan before implementation begins. Users can reword, reorder, remove, or add steps first.
Aperture
Freemium
An AI code editor that plans changes, stages diffs, and checks them before application.
What is Aperture?
Aperture is an AI code editor built around a plan-before-apply workflow. Composer reads a codebase, proposes a plan, and waits for the user to start the build. Users can revise the plan by rewording, reordering, removing, or adding steps before any edit is made.
The editor delivers proposed changes as staged diffs rather than writing directly to files. Users can keep or skip changes file by file, while each run groups its plan, build action, test runs, and automatic fixes in one place. Aperture checks staged changes before application by parsing them, resolving imports, running a light type check, rendering the page, and executing tests.
The repository also includes a replayable harbor-api demo project. With replay mode enabled, recorded agent runs can be viewed without calling a model or using an API key. The documented local setup requires Node 22.6 or newer.
What can Aperture do?
Staged diffs instead of immediate edits
Changes arrive as staged diffs that can be kept or skipped file by file, so edits do not touch project files until the user applies them.
Pre-apply verification
Each staged change is checked for parsing errors, unresolved imports, light type-check failures, page-rendering problems, and test failures before application.
Run cards and background work
A run card combines the plan, build action, test runs, and automatic fixes, with background jobs shown alongside active work.
Side-by-side change comparison
Two runs with staged changes can be compared file by file using the documented Ctrl/Cmd+J shortcut.
Replayable demo mode
APERTURE_MODEL=replay plays back recorded agent runs on the built-in harbor-api demo without calling a model or consuming an API quota.
Use Cases
“Review an AI-proposed code change”
Ask Composer to plan a change, adjust the steps, and inspect the resulting diff before applying anything to the project.
“Validate a bug fix in a demo project”
Use the harbor-api demo and its recorded runs to watch a proposed fix move through planning, staged changes, rendering, and tests.
“Separate new regressions from existing failures”
When a project has known failing tests, use the test result context to see whether a failure was already present before the staged edit.
“Compare alternative agent runs”
Keep multiple staged runs available and compare their files side by side before choosing which change set to apply.
Frequently Asked Questions
Does Aperture write changes directly to project files?
No. Proposed edits arrive as staged diffs that can be kept or skipped file by file. Files are not changed until the user applies the selected edits.
What does Aperture check before a change is applied?
The documented checks cover parsing, import resolution, a light type check, page rendering, and tests.
Can Aperture be tried without an API key?
Yes. The repository documents APERTURE_MODEL=replay, which plays recorded agent runs on the built-in harbor-api demo without calling a model or using an API quota.
What is required for the local setup?
The documented setup requires Node 22.6 or newer. The repository also provides a VITE_AUTH_ENABLED=false option that skips sign-in and uses a local in-process PGlite database.
Quick Facts
- Category
- AI developer tool
- Primary workflow
- Plan, stage a diff, verify, then apply
- Verification
- Parsing, imports, light type check, page rendering, and tests
- Local requirement
- Node 22.6 or newer
- Demo mode
- Recorded runs for the built-in harbor-api project
- Source
- github.com/Hankaws/aperture
Aperture Alternatives
PearAI
trypear.ai
Open-source AI code editor with chat, agentic coding, and model routing, plus free, Maker, and Enterprise plans.
ponytail
ponytail.dev
ponytail is a ruleset for AI coding agents that guides them toward the smallest code change that works. It helps developers and agent users avoid overengineering, reuse platform features, and keep changes concise.
Diffsmith
apps.mergeable.io
Diffsmith is a native macOS app for reviewing local git changes made by AI coding agents such as Claude Code, Cursor, Codex, and Copilot. Add inline comments and send them back to the agent as a prompt or via a local MCP server.
AutonomyAI
autonomyai.io
AutonomyAI is an AI-native product delivery platform that helps product managers and designers turn ideas into production-ready code changes in their existing codebase. Engineers review and approve the resulting pull requests before merge.
Factory
factory.ai
Factory's Droids inspect codebases, propose scoped changes, and show diffs for review before editing.
Onlook
onlook.com
Onlook is a visual code editor for designers and product teams, with live React editing, AI-assisted changes, and self-hosting.