Expand description
greplm-core: an extreme-performance, trigram-based code indexer for LLM agents.
The index lives in a .greplm/ directory at the project root and consists of
immutable, mmap-backed segments (trigram FST + roaring posting lists + doc and
symbol tables). Search filters candidate documents by trigram intersection,
then verifies matches with the real literal/regex matcher.
Modules§
- cache
- Change-detection cache backed by redb.
- client
- Client for the greplm daemon.
- config
- Index configuration persisted at
.greplm/config.toml. - context
- Task-driven context packs.
- daemon
- Warm-index daemon.
- git
- Git time-travel intelligence.
- indexer
- Index construction: full builds, hash-gated incremental updates, compaction.
- io_
backend - Ingest read path.
- lang
- Language detection and tree-sitter grammar wiring.
- meta
- Index manifest persisted at
.greplm/meta.json. - paths
- proto
- Wire protocol shared by the daemon and its clients.
- resolve
- Identifier resolution: turn a source position into the identifier under the cursor, classify how it is used (call, method/member access, type position, import), and provide per-language hints that bias definition ranking.
- savings
- Token-savings accounting for agent queries.
- search
- Query execution: trigram candidate filtering, then exact verification.
- segment
- On-disk index segments.
- structural
- Structural (AST) search.
- symbol
- Symbol extraction via tree-sitter.
- trigram
- Trigram extraction and query decomposition.
- walk
- Filesystem traversal.
- watch
- Live incremental indexing driven by filesystem events.
Structs§
- Greplm
- A handle to a greplm project (a directory and its
.greplmindex). - Status
- Status snapshot for
greplm status.
Enums§
- Error
- Errors produced by the greplm core engine.
Type Aliases§
- Result
- Result type used across greplm-core.