Skip to content

cli.md

Latest commit

 

History

History
111 lines (82 loc) · 4.29 KB

cli.md

File metadata and controls

111 lines (82 loc) · 4.29 KB

CLI reference

The redcell command is a Typer app (redcell/cli.py, entry point redcell = redcell.cli:app). Run any command with uv run redcell <command> (or redcell <command> in an activated venv).

redcell version          # print the installed version
redcell chat             # interactive REPL against a configured model
redcell serve            # OpenAI-compatible HTTP server (+ AgentGateway)
redcell rag-seed         # load the RAG corpus into Qdrant via the gateway

All commands read configuration from env / .env (see configuration.md).

redcell version

Prints the package version (redcell.__version__). No options.

redcell chat

Starts an interactive REPL. Each line you type is one Agent.run() turn against the agent's own in-memory history, so the conversation accumulates within the session. Tool calls (builtin tools only — chat does not launch the gateway) run as needed.

uv run redcell chat
uv run redcell chat --system "You are a terse pentest assistant."
Option Default Meaning
--system <text> You are a helpful assistant. Base system prompt. The safety policy is still prepended on top of it when AGENT_SAFETY_PROMPT=true.

The active guardrail (AGENT_GUARDRAILS) applies here too. Exit with Ctrl-C.

redcell serve

Runs the OpenAI-compatible HTTP API (see server-api.md) and, unless disabled, launches and supervises an AgentGateway process so MCP tools are available.

uv run redcell serve
uv run redcell serve --host 127.0.0.1 --port 9000
Option Default Meaning
--host <host> AGENT_SERVER_HOST (0.0.0.0) Override bind host.
--port <port> AGENT_SERVER_PORT (8800) Override bind port.

On start it prints the bind URLs and the active security posture, e.g.:

Serving agent (anthropic/claude-opus-4-8) as model 'redcell'.
  security: safety_prompt=on, guardrails=on
  local:  http://127.0.0.1:8800/v1
  docker: http://host.docker.internal:8800/v1  (use this in Open WebUI)
  gateway: launching 'agentgateway' on :3030
  qdrant:  docker compose up -d qdrant (RAG store on :6333)
  docs:    ingesting PDFs from 'documents/' into the RAG store

If AGENT_MCP_TOOL_DENYLIST is set, the denied tool names are printed too.

Gateway behavior: controlled by AGENT_GATEWAY_* (see configuration.md). If the binary is missing or never becomes ready, serve logs a warning and runs with builtin tools only — it does not fail. Set AGENT_GATEWAY_AUTOSTART=false to run the gateway yourself.

Qdrant (RAG store): controlled by AGENT_QDRANT_*. serve brings Qdrant up via docker compose up -d (started before the gateway so the rag target finds a store) and waits for its port. Needs Docker; if absent it logs a warning and runs without RAG. Left running on exit by default — set AGENT_QDRANT_STOP_ON_EXIT=true to stop it, or AGENT_QDRANT_AUTOSTART=false to manage it yourself. See rag.md.

Document ingestion: controlled by AGENT_DOCS_*. After the gateway/Qdrant are up, PDFs in AGENT_DOCS_DIR (default documents/) are chunked and stored into the RAG store in the background, so the agent can retrieve them via qdrant-find. A hash manifest skips unchanged files across restarts. Disable with AGENT_DOCS_AUTOLOAD=false. See rag.md.

redcell rag-seed

Loads a corpus of documents into the RAG vector store through the gateway's qdrant-store tool (so embeddings/collection match what qdrant-find retrieves). Requires a running gateway with the rag target and a running Qdrant. See rag.md.

docker compose up -d qdrant     # start Qdrant
uv run redcell serve            # in another shell — brings up the gateway + rag target
uv run redcell rag-seed         # load the bundled corpus
uv run redcell rag-seed --corpus path/to/your_corpus.json
Option Default Meaning
--corpus <path> bundled seed_corpus.json Corpus JSON file to load.

Prints the number of documents stored. Fails loudly if no qdrant-store tool is reachable or if any store call errors (e.g. Qdrant down).