Skip to content

README.md

Latest commit

 

History

History
79 lines (57 loc) · 3 KB

README.md

File metadata and controls

79 lines (57 loc) · 3 KB
Raw

@nodetool-ai/cli

Command-line tools for NodeTool — interactive AI chat and workflow management from the terminal.

Ships two executables: nodetool for managing workflows, jobs, assets, models, secrets, and servers, and nodetool-chat for an interactive multi-provider LLM chat with tool support.

Install

npm install -g @nodetool-ai/cli

Executables

Command Description
nodetool Management CLI — serve, run workflows, jobs, assets, models, secrets, settings, MCP install
nodetool-chat Interactive terminal chat across providers

nodetool

nodetool info                       # System and environment info, API-key status
nodetool serve --host 0.0.0.0 --port 7777   # Start the WebSocket + HTTP server
nodetool run workflow.ts            # Run a TypeScript/JavaScript DSL workflow file

# Workflows (id-based commands read the local database; --api-url for a remote server)
nodetool workflows list
nodetool workflows get <id>
nodetool workflows run <id|file.json|file.ts> --params '{"key":"value"}'
nodetool workflows export-dsl <id> -o out.ts       # Export as TypeScript DSL
nodetool workflows export-bundle <id> -o pack.nodetool   # Portable zip with assets
nodetool workflows import-bundle pack.nodetool

# Jobs and assets
nodetool jobs list --workflow-id <id>
nodetool jobs get <job_id>
nodetool assets list --query "photo" --content-type image/png

# Models
nodetool models list
nodetool models providers
nodetool models by-provider anthropic --kind llm

# Secrets (encrypted local store)
nodetool secrets list
nodetool secrets store OPENAI_API_KEY
nodetool secrets get OPENAI_API_KEY

# Settings and MCP
nodetool settings show
nodetool mcp install                # Register the MCP server for Claude Code, Codex, OpenCode
nodetool mcp status

The workflows, jobs, assets, and models read commands hit the local database, providers, and caches by default. Pass --api-url <url> (env NODETOOL_API_URL) to query a remote server instead, and --json for machine-readable output.

nodetool-chat

nodetool-chat                           # Saved / auto-detected settings
nodetool-chat --provider anthropic --model claude-sonnet-5
nodetool-chat --workspace /path/to/dir  # Set the workspace for file tools
nodetool-chat --url ws://localhost:7777/ws   # Connect to a running server
echo "research 5 AI topics" | nodetool-chat   # Piped, non-interactive

Chat flags: -p, --provider, -m, --model, -u, --url, -w, --workspace, --tools <list>. Every session runs the unified agent loop; -a, --agent and --no-agent are accepted for backwards compatibility and do nothing.

Providers include anthropic, openai, gemini, xai, groq, mistral, deepseek, ollama, lmstudio, and other registry providers.

Tracing

Both executables accept --trace-file <path> (append LLM/agent/workflow spans as JSONL) and --trace-stdout <pretty|json>.

Links