Lexa

Lexa includes deterministic agent tool-suite benchmarks over a synthetic project with known ground truth. Token counts estimate the default structured-text CLI and MCP output.

cargo test --test agent_retrieval_benchmark -- --nocapture
cargo test --test agent_cli_output_format_benchmark -- --nocapture
cargo test --test agent_edit_safety_benchmark -- --nocapture
cargo test --test agent_mcp_session_benchmark -- --nocapture
cargo test --test agent_maintenance_benchmark -- --nocapture
Latest local runJune 30, 2026
Correctness checks34/34
Retrieval aggregate reduction80.0%
CLI output format1676 est. tokens across 13 commands
retrieval13/13

Indexed answers against listing, grep, and candidate reads.

Lexa
1678
Baseline
8397
Reduction
80.0%
cli_output_format13 cases

Current structured-text CLI envelope across retrieval commands.

Lexa
1676
Baseline
n/a
Reduction
n/a
edit_safety11/11

Hash-aware reads, dry runs, patches, creates, and rejection paths.

Lexa
709
Baseline
n/a
Reduction
n/a
mcp_session5/5

Persistent MCP session edits and state queries.

Lexa
281
Baseline
n/a
Reduction
n/a
maintenance5/5

Recent files, status, reindex, audit, and clear-index.

Lexa
611
Baseline
n/a
Reduction
n/a

Retrieval is the suite with a full non-indexed baseline: recursive file listing, grep-style search, and candidate file reads. Safety and state suites report correctness and token envelope only.

TaskToolLexaBaselineReductionCorrect
filtered file overviewfiles8532-165.6%true
directory childrenlist10252-96.2%true
glob pathsglob3020-50.0%true
fuzzy pathpath_search2812978.3%true
scoped text searchtext_search177105-68.6%true
exact word refsword_refs304274-10.9%true
exact definitionsymbol_defs48176897.3%true
fuzzy symbolsymbol_search86146294.1%true
callerscallers33094565.1%true
outlineoutline1041072.8%true
dependenciestrace_deps3211371.7%true
briefbrief220244591.0%true
composed querypipeline13294586.0%true