RXP CLI Reference
countersignal rxp [OPTIONS] COMMAND [ARGS]
RAG retrieval poisoning optimizer. Validate whether poison documents achieve retrieval in vector similarity searches across embedding models.
Commands
| Command | Description |
|---|
list-models | Show registered embedding models |
list-profiles | Show built-in domain profiles |
validate | Run retrieval validation against a corpus |
list-models
Show all registered embedding models with their HuggingFace model names and vector dimensions.
countersignal rxp list-models
No options. Prints a table:
| ID | Model | Dimensions | Description |
|---|
minilm-l6 | sentence-transformers/all-MiniLM-L6-v2 | 384 | Open WebUI default embedding model |
minilm-l12 | sentence-transformers/all-MiniLM-L12-v2 | 384 | Higher quality MiniLM variant |
bge-small | BAAI/bge-small-en-v1.5 | 384 | BGE small English v1.5 |
This command does not require the optional RXP dependencies (sentence-transformers, chromadb).
list-profiles
Show built-in domain profiles with query and corpus document counts.
countersignal rxp list-profiles
No options. Prints a table:
| ID | Name | Queries | Corpus Docs |
|---|
hr-policy | HR Policy Knowledge Base | 8 | 5 |
This command does not require the optional RXP dependencies (sentence-transformers, chromadb).
validate
Run retrieval validation — embed a corpus, ingest into ChromaDB, query, and score poison document retrieval.
countersignal rxp validate [OPTIONS]
Options
| Option | Type | Default | Description |
|---|
--profile | TEXT | — | Domain profile ID (e.g. hr-policy) |
--corpus-dir | PATH | — | Path to custom corpus directory |
--poison-file | PATH | — | Path to poison document |
--model | TEXT | minilm-l6 | Embedding model: registry shortcut, all, or HuggingFace model name |
--top-k | INTEGER | 5 | Number of retrieval results per query |
--output | PATH | — | Write JSON results to file |
--verbose / -v | — | off | Show per-query hit details |
Either --profile or --corpus-dir is required. The command exits with an error if neither is provided.
Using a Built-in Profile
Run validation using the hr-policy profile with the default embedding model:
countersignal rxp validate --profile hr-policy
Run with verbose output to see per-query retrieval details:
countersignal rxp validate --profile hr-policy --verbose
Comparing Across Models
Use --model all to run validation against every registered embedding model and get a comparison table:
countersignal rxp validate --profile hr-policy --model all
Using a Specific Model
Use a registry shortcut:
countersignal rxp validate --profile hr-policy --model bge-small
Or pass any HuggingFace model name directly:
countersignal rxp validate --profile hr-policy --model BAAI/bge-m3
Custom Corpus and Poison Files
Provide your own corpus directory (all *.txt files will be loaded) and poison document:
countersignal rxp validate \
--corpus-dir ./my-corpus \
--poison-file ./my-poison.txt \
--model minilm-l6
JSON Output
Save results to a JSON file for programmatic analysis:
countersignal rxp validate --profile hr-policy --model all --output results.json