-
Notifications
You must be signed in to change notification settings - Fork 0
Command search
Natural-language semantic search across code and durable knowledge.
columbus search <query> [flags]The query is embedded on-device and matched by vector similarity (k-nearest
neighbors over the vec0 store), then re-ranked by deterministic heuristics —
finding results by meaning, not just exact words. When no onnxruntime runtime
is available it degrades to keyword (FTS) ranking; columbus doctor shows which.
Results span code, free-form memory, and the epic → story → task hierarchy.
Every hit carries a score and a human-readable why (semantic match for
vector hits, or a heuristic reason). Search is locate-first: by default it
returns locations, signatures and scores — add --snippets for code bodies. For
usage patterns see Searching Effectively.
| Flag | Default | Description |
|---|---|---|
--kind string |
all |
What to search: code|memory|epic|task|all
|
--limit int |
15 |
Maximum number of results (global flag) |
--context-lines int |
3 |
Lines of context around matched ranges |
--snippets |
off | Attach code bodies inline |
--snippet-lines int |
0 |
Cap snippet length in lines (0 = default 60) |
Plus global flags (--json, --llm, --no-color). For the dependency graph,
see Command: graphs.
$ columbus search "how are credentials checked"
1. Check [method] (score 0.51, semantic match)
internal/auth/auth.go:9
2. Authenticator [class] (score 0.46, semantic match)
internal/auth/auth.go:6
...
$ columbus search "token validation" --kind all --limit 5
# code symbols + any matching epics/stories/tasks/memories, ranked together{"command":"search","query":"how are credentials checked","kind":"all","total":3,"ok":true,"schema_version":1,
"hits":[{"grain":"symbol","name":"Check","symbol_kind":"method",
"path":"internal/auth/auth.go","start_line":9,"end_line":9,
"score":0.5081,"why":"semantic match","risk_level":"low","role":"impl","exported":true}, …]}0 success · 3 not initialized / index missing. See Exit Codes.
Searching Effectively · Navigating Code · Command: graphs · Output Modes
Columbus — the navigator your coding agent has been missing · local-only, deterministic code context · Repository · Issues · MIT License
Getting started
Concepts
Guides
- Using Columbus with Your Agent
- Searching Effectively
- Navigating Code
- Project Memory
- Tracking Work: Epics, Stories & Tasks
- Keeping the Index Fresh
Command reference
Reference
- Output Modes
- JSON Contract & Errors
- Exit Codes
- Configuration
- Supported Languages
- Color & Environment
Project