Configuration
Configuration is a HOCON file; start from
keyvector.example.conf.
Every value can be overridden by an environment variable named after its path, without the
keyvector. prefix, upper-cased, with dots and dashes as underscores. For example,
keyvector.embedder.url becomes EMBEDDER_URL. The CloudFormation template configures the
container entirely this way.
| Setting | Default | Meaning |
|---|---|---|
stream.keyspace-name, stream.table-name |
The table to sync. | |
stream.application-name |
Names the KCL lease tables. One per table; changing it starts the stream over. | |
sync.text-columns |
Columns joined (blank line between) as the text to embed. | |
sync.text-template |
Or a template, e.g. "{title}: {overview}". Set exactly one of the two. |
|
sync.max-text-chars |
8000 |
Text is cut to this length. |
sync.metadata-columns |
[] |
Columns stored with each vector, for filtering. |
sync.store-text-as |
Also store the embedded text under this metadata key (make it non-filterable). | |
sync.on-empty-text |
delete |
delete the vector or skip when a row’s text becomes empty. |
sync.key-columns |
Only if stream records carry no primary key (they normally do). | |
embedder.type |
bedrock |
tei, bedrock or openai. |
embedder.model |
Titan V2 | Bedrock or OpenAI model id. |
embedder.dimensions |
Must equal the model’s output and the index. | |
embedder.url |
TEI base URL, or an OpenAI-compatible base URL. | |
embedder.api-key-env |
Name of the environment variable holding an API key; never the key itself. | |
embedder.batch-size |
32 |
Texts per request (Titan: always 1; Cohere: at most 96). |
embedder.concurrency |
4 |
Requests in flight. |
embedder.requests-per-minute |
0 (Titan: 60) |
Most requests started per minute; 0 means no limit. |
embedder.document-prefix, query-prefix |
For models that want them. | |
sink.vector-bucket, sink.vector-index |
The S3 Vectors index. | |
processor.max-attempts |
10 |
After this many failures a batch goes to the dead-letter queue, if set. |
processor.dead-letter-queue-url |
SQS queue for batches that keep failing. Without one, the shard keeps retrying. |