Troubleshooting¶
Start with the command that best matches the failure:
kbolt doctorfor setup and readiness problemskbolt local statusfor managed local-service problemskbolt watch statusfor automatic freshness problemskbolt statusfor index and storage state
doctor says kbolt is not set up¶
On a fresh machine, doctor exits with:
That means the config has not been created yet. Run:
setup local fails¶
Check these first:
llama-serveris installed and on yourPATH- the machine can download the default model files
- local ports for the managed services are available
Then rerun:
If setup completes but later readiness checks fail, inspect:
If you installed with Homebrew, llama-server should come from the llama.cpp dependency. For Cargo or GitHub Release installs, install llama-server separately.
doctor warns that the index does not exist yet¶
If you already ran setup local but have not indexed any collection yet, storage warnings are expected. The config and model stack can be healthy before any index files exist.
Create the index by adding a collection or running an update:
or:
Local services are configured but not reachable¶
You may see error text like:
or a provider transport error that includes a local URL such as http://127.0.0.1:8101.
Run:
Then try:
If the services still do not stay up, inspect the log files under the kbolt cache directory. See Data locations.
Command says no active space¶
You may see:
Fix it by choosing one of these:
Use kbolt space list if you are not sure which spaces exist.
Search returns no results¶
Check these in order:
- the collection exists
- the collection was indexed successfully
- you are searching the intended space
- the query mode matches the query shape
Useful commands:
If files changed after the first indexing pass, make sure the watcher is running:
If you need a refresh immediately, run:
Changed files do not appear in search¶
Check whether automatic watching is enabled and running:
On macOS and Linux, if the watcher is disabled, start it:
If the watcher reports an error, inspect recent logs:
If you need fresh keyword and metadata results immediately, run:
If semantic search is stale while a file is actively changing, wait for the collection to become quiet or run a full update:
If watch status says semantic indexing is blocked for a space, run the exact kbolt --space <space> update command shown in the status output.
Indexing is slower than expected¶
The most common cause is indexing too much irrelevant material.
Check:
- collection-local
.gitignore - default ignored directories
- custom
kbolt ignorerules
Use:
Then tighten the exclusion set. If the watcher is running, it will refresh the affected collection automatically; run kbolt update when you want the change applied immediately.
Update output says more errors were omitted¶
When update or initial indexing has many file errors, the default output shows a short list and a hint such as:
Rerun the update with --verbose:
If the failed collection was scoped, keep the same --space and --collection values in the verbose command.
Still stuck?¶
- Re-run Quickstart against a small test directory first.
- Use Health and status to choose the right status command.
- Compare the current paths with Data locations.
- Check automatic freshness in Keep indexes fresh.
- Use Search modes if the problem is retrieval quality rather than availability.