-
Notifications
You must be signed in to change notification settings - Fork 1
How to keep the index fresh
Arun Soman edited this page Aug 31, 2026
·
1 revision
cie index . # 110 files / 1,900+ nodes in ~1.9s; writes .cie/graph.db-
Single file: the
reindex_filetool (orcie reindex <path>) re-extracts just that file and re-resolves its call edges in place. -
Hands-off:
cie watch(watchdog) auto-reindexes on file events; thestart_watch/stop_watchMCP tools do it live from a client. -
Bulk: re-run
cie index .— it's a full replace, and fast.
The freshness_report tool (ToolService) returns per-file index age
and missing/changed files since the last index — an honest answer, not
a guess. If a file changed, its stale data is reported, not silently
served as fresh.
- Deleted
.cie/? Just re-index — nothing else is needed. - Neo4j mode: freshness is per
--projectnamespace. -
cie indexruns the full pipeline: extract → call resolution → TESTS pass (see Extraction pipeline).
If code and this wiki disagree, the code wins — then this wiki gets a PR. Evidence lives in the repo, not here.
Start
How-tos
- Install & serve to your MCP client
- Choose your storage backend
- Run the Neo4j team mode
- Index & keep it fresh
- Ask impact questions
- Task & QA layer
- Semantic search
- Snapshot or serve HTTP
- Policies & the write boundary
- Troubleshoot
Reference
- Extraction pipeline
- Data model
- Tool reference
- Benchmarks
- Security & determinism
- Two tiers · Project layout · Docs index
Contribute