Compare
ashlr vs the alternatives
An honest comparison with WOZCODE, native Claude Code, Codex, and Cursor. Numbers for ashlr come from the reproducible benchmark. Numbers marked unknown are not publicly available for that product — we will not invent them.
Last updated April 2026 · Cursor pricing from cursor.com
+ yes / included— nopartialunknown (not published)
| Feature | ashlr | WOZCODE | Native Claude Code | Native Codex | Cursor |
|---|---|---|---|---|---|
| Compression | |||||
Read compression ashlr number from docs/benchmarks-v2.json; others not publicly benchmarked | −82.2% mean | unknown | none | none | unknown |
Grep compression ashlr no-genome baseline; genome path higher in practice | −81.7% mean | unknown | none | none | unknown |
Edit token overhead Small edits: ashlr diff-summary adds overhead vs native | −52% medium, −96.5% large | unknown | none | none | unknown |
| Architecture | |||||
Multi-file atomic edit ashlr__multi_edit: rollback on failure across N files in one call | + | unknown | — | partial | partial |
Shared genome / retrieval index ashlr genome is TF-IDF + optional Ollama semantic; free tier is local only | + | — | — | — | — |
Auto-refresh genome on edits _genome-live.ts patches genome sections after every ashlr__edit | + | — | — | — | — |
| Observability | |||||
Per-session token ledger | + | unknown | — | — | — |
Status-line visibility Animated sparkline in Claude Code terminal footer | + | — | — | — | — |
Real-time counters Worst-case latency ~550 ms; mtime-invalidated cache | + | — | — | — | — |
Public reproducible benchmark scripts/run-benchmark.ts; weekly CI refresh; docs/benchmarks-v2.json | + | — | — | — | — |
| Cloud & sync | |||||
Cross-machine stats sync | Pro+ | unknown | — | — | — |
Hosted LLM summarizer ashlr Pro routes to hosted Haiku-4.5 endpoint; free tier uses local Ollama or snipCompact fallback | Pro+ | unknown | — | — | — |
Team shared genome (E2E encrypted, vclock-merged) | Team+ | — | — | — | — |
| Open source & portability | |||||
Open source Claude Code core is proprietary; native tools are not open source | MIT | — | partial | partial | — |
Self-hostable Full plugin + backend deployable on-prem; genome format is a public spec | + | — | — | — | — |
Data residency control ashlr free tier: nothing leaves the machine. Enterprise: on-prem + private inference. | + | — | — | partial | partial |
| Compatibility | |||||
MCP compatibility | + | unknown | + | + | + |
Cursor compatibility ashlr MCP servers run under Cursor; Codex and Claude Code get the richer plugin surfaces | + | unknown | — | + | + |
Codex compatibility Ashlr ships .codex-plugin, .mcp.json, Codex skills, and nudge-first hooks | + | unknown | — | + | — |
Goose compatibility ports/goose/recipe.yaml ships with the plugin | + | unknown | — | — | — |
| Pricing & team | |||||
Pricing Cursor pricing from cursor.com as of April 2026; subject to change | Free · $12/mo Pro · $24/seat/mo Team | unknown | Included with Claude Code | Included with Codex | Free · $20/mo Pro · $40/seat/mo Business |
Team features | Team tier | unknown | — | — | Business tier |
Methodology note
ashlr numbers are from scripts/run-benchmark.ts run against the plugin's own repository (337 files, 56,901 LOC) at commit 7f63e08. The benchmark samples four file-size buckets deterministically seeded from the commit SHA. Token counts use the chars/4 heuristic used at runtime. Results for other products are either not published or not independently reproducible; they are marked “unknown.”