Aveyx is an offline tool that gives AI coding assistants a slimmer view of your code
A developer built a free, offline tool that shrinks code for AI assistants and previews what a change could break before you merge it.
The short version
- Aveyx v3.0.4 runs offline and works as a zero-config MCP server over stdio
- 'aveyx compact' strips function bodies but keeps types, signatures and key callers
- 'aveyx blast-radius' checks what a change can reach before you merge
- 'aveyx vault' makes encrypted snapshots with 1-click rollbacks
- MIT licensed, 144/144 tests passing, 0 CodeQL alerts, per the author
Why it matters
AI coding assistants tend to struggle once a project gets large: they get swamped with text, make edits without seeing the wider effects, and sometimes break things quietly. Aveyx is one developer's answer, and its design is worth studying even if you never install it. Running locally also means your code does not have to leave your machine.
The problem Aveyx targets
As dev.to reported, the author says AI coding assistants such as Cursor, Claude Code, Windsurf, Copilot and Antigravity handle small, isolated edits well. Trouble starts on a real repository, where the author lists four failure modes: context bloat, blind refactors, silent regressions and irreversible mistakes.
Context bloat means the assistant's working memory fills up with code it does not need. A blind refactor is a restructuring done without knowing what else depends on the code being changed.
What the tool does
Aveyx (version 3.0.4) bundles four jobs into one offline package. It also runs as an MCP server, meaning it follows a standard that lets AI tools call outside programs, and it needs no configuration.
- compact uses a Babel AST (a structured map of source code) to strip out function bodies while keeping types, signatures and the callers ranked by PageRank, the link-ranking idea made famous by web search.
- refactor makes edits that take the code's dependency graph into account and shows an impact preview first.
- blast-radius runs a reachability analysis, tracing what a change could affect, before you merge.
- vault keeps encrypted snapshots and allows one-click rollbacks.
What the author claims, and what we cannot check
The project is MIT licensed, a permissive open-source license. The author reports 144 of 144 tests passing and zero CodeQL alerts, which refers to GitHub's automated code-security scanner.
These are self-reported figures from a launch post. The source gives no measurements of how much context the compaction saves or how accurate the blast-radius analysis is, so those claims remain untested here.
The idea worth borrowing
The useful pattern is simple: instead of pasting whole files into an assistant, give it a compact outline and a way to ask what a change touches. Keeping the work local is a bonus for anyone wary of sending code elsewhere.
What to watch
- Whether independent users report real context savings from the compaction
- How accurate the blast-radius analysis proves on large, messy codebases
Try it
If you use an AI coding assistant on a large project, check the GitHub repo and npm page for Aveyx and try it on a throwaway branch first, comparing its compact outline with what you normally paste in.
Sources and links
- Reportingdev.to: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by dev.to · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
