lf chat
Send a single prompt to your project’s chat endpoint. By default, responses include RAG context defined in llamafarm.yaml.
Synopsis
lf chat # Interactive TUI using project from llamafarm.yaml
lf chat [namespace/project] # Interactive TUI for explicit project
lf chat [namespace/project] "message" [flags] # One-off request
If you omit namespace/project, the CLI resolves them from llamafarm.yaml.
Useful Flags
| Flag | Description |
|---|---|
--model | Select a specific model from your multi-model configuration. |
--file, -f | Read prompt content from a file. |
--no-rag | Skip retrieval—direct LLM call. |
--database | Target a specific RAG database. |
--retrieval-strategy | Override the retrieval strategy. |
--rag-top-k | Adjust the number of results (default 5). |
--rag-score-threshold | Minimum similarity score for results. |
--curl | Print the sanitized curl request instead of executing. |
Behaviour
- Automatically starts the server if needed.
- Filters client/error messages from the transcript before sending.
- Streams responses; exit code is non-zero if the API returns an error.
- Redacts authorization headers when using
--curl.
Examples
# Interactive project chat (auto-detect project)
lf chat
# Basic one-off chat (RAG enabled)
lf chat "Summarize the latest FDA letters."
# Use a specific model
lf chat --model powerful "Complex reasoning question"
# Explicit project with file input
lf chat company/legal -f prompt.txt
# Pure LLM request with curl preview
lf chat --no-rag --curl "Explain RAG in 2 sentences"
# Override strategy for targeted retrieval
lf chat --database main_db --retrieval-strategy hybrid_search "Find biologics references"
# Combine model selection with RAG
lf chat --model lemon --database main_db "Query with specific model and database"
Sessions
- Set
LLAMAFARM_SESSION_ID=abc123to keep context between calls. lf startmanages its own session history in.llamafarm/projects/.../dev/context.- Delete session files to reset state or start a new namespace/project for isolation.