ig instant-grep
Head-to-Head Comparison -- v1.10.0

ig vs RTK -- Head to Head

115 cases across 12 domains, run on the iautos SaaS monorepo (347,843 files raw, 3,146 after ig's default excludes). 2 warm-up passes + median of 3 wall-time runs per case. ig 1.10.0 against rtk 0.37.2. Bytes measured with wc -c.

The Verdict

115
Benchmark cases across 12 domains (v1.10.0)
896K / 1.04M
ig total bytes vs rtk — ig wins
1.74s / 2.88s
ig total time vs rtk — ig 40% faster
15 / 15
--top + --semantic bytes wins (rtk: 0)

First release where ig beats rtk on aggregate bytes and aggregate wall time simultaneously. The two new v1.10.0 capabilities (--top N BM25, --semantic PMI) account for 15 of ig's 57 bytes wins — they are structural advantages rtk cannot close without building its own persistent index.

Totals -- 115 cases

ig 1.10.0 rtk 0.37.2
Total bytes emitted 896 KB 1.04 MB
Total wall time 1.74 s 2.88 s
Bytes wins (tie: 4) 57 / 115 54 / 115
Time wins (tie: 8) 80 / 115 27 / 115

Raw CSV + markdown report: results.csv, summary.md, per-domain.md.

Per-Domain Breakdown

# Domain ig B wins rtk B wins ig T wins rtk T wins
1 literal search 5 5 9 1
2 regex search 3 7 6 4
3 flag variants 7 3 9 1
4 listing 2 8 7 3
5 read full 0 10 8 0
6 read signatures 9 1 10 0
7 git proxy 7 2 8 0
8 varied identifiers 3 4 10 0
9 smart summaries 4 6 0 10
10 generic proxy 2 8 4 3
11 --top BM25 moat 10 0 7 3
12 --semantic PMI moat 5 0 2 2

The Index Moat

rtk is a generic output-compression proxy: it shells to rg, git, find... and compresses their output. ig is a search engine with a persistent index — which unlocks capabilities rtk cannot match without completely redesigning itself.

--top N (BM25 ranking)

Score each matched file by Okapi BM25 and return only the top N. rtk has no tf / df / avdl so it cannot rank — it can only cap flat output length. 

ig --top 10 "export default"
  →  743 B

rtk grep "export default"
  →  19 403 B   (96% more)

--semantic (PMI expansion)

Synonyms learned from the corpus at index time (count-weighted PPMI), zero ML model. rtk has no co-occurrence store; there's nothing to expand against. 

ig --semantic --top 5 throw
  → got, inattendu, denied,
     autorisé, trouvée, manquant
  →  3 368 B

rtk grep throw
  →  17 717 B

Where rtk Still Wins -- and Why It's a Trade-off

  • Read full (10/10 bytes to rtk). ig keeps the 42: content line-number prefix because it's what lets the Edit tool round-trip precisely. Dropping it saves ~15% bytes per file and halves the utility. Deliberate trade-off, not a gap.
  • Listing singles (8/10 bytes). rtk's rtk ls emits a placeholder 8 B for top-level dirs. Fewer bytes, less information; ig emits a compact listing that's still actionable.
  • Smart per-file (6/4 bytes). Byte difference is 50-150 B per case — noise-level. In aggregate, ig's dir-aggregate mode is already 200× smaller than the legacy per-file walk (see cli docs).

Feature Comparison

Feature ig rtk
BM25 --top N ranking moat v1.10.0 No (no index)
--semantic PMI query expansion moat v1.10.0 No (no index)
Auto-compact on pipe v1.10.0 Always
Path ellision in compact headers v1.10.0 No
Dir-aggregate smart summaries v1.10.0 No
Trigram-indexed search moat Yes No (wraps rg)
Sub-ms Unix-socket daemon moat Yes No
IDF-weighted sparse n-grams moat Yes No
Bloom + locMask (3.5-gram) moat Yes No
Pre-computed filedata cache moat Yes No
Budget mode (entropy-scored) moat Yes No
Relevance boost (-r) moat Yes No
Delta reads (git changes) moat Yes No
Token savings dashboard ig gain (top-20 default) rtk gain
Git output compression Yes Yes
ls / tree compression Yes Yes
File reading modes moat 4 modes 3 modes
Pipelines in rewrite Yes Yes
Env / sudo prefix stripping Yes Yes
Absolute path normalisation Yes Yes
Git global opts (-C, --git-dir) Yes Yes
Dedup consecutive output Yes Yes
Rewritten command categories moat 91 bins 72 rules

moat rows mark structural advantages — they require a persistent index, which rtk does not have.

ig gain Dashboard

Real output from ig gain (v1.10.0 defaults to showing the top 20 commands; --full for the complete list).

ig gain 
ig Token Savings
================================================================

Total commands:    3,445
Input bytes:       28.7M
Output bytes:      21.4M
Bytes saved:       7.3M (25.4%)

By Command  (top 20)
-----------------------------------------------------------------------
  #   Command                        Count    Saved    Avg%  Impact
-----------------------------------------------------------------------
  1.  ig read -s                       343     3.9M  92.7%  ==========
  2.  ig git diff                      184     1.9M  70.6%  =====
  3.  ig run curl                       30   716.0K  92.1%  ==
  4.  ig run cargo test                  43   555.9K  76.1%  =
  5.  ig git log                       145   476.6K  46.2%  =
  6.  ig ls                            491   358.1K  77.2%  =
  …
 20.  ig run /tmp/bench/mocks...        14    24.5K  29.3%
-----------------------------------------------------------------------
Showing top 20 of 143 commands. Use `ig gain --full` to see the full list.

Reproduce

The full script, the 115-case CSV, and the per-domain markdown report are published alongside this release: