116 online 3 members  + Join RP Base  +
FreeTool

venice-cli

by veniceai  ·  version 2026.11.13  ·  updated yesterday

No reviews yet

Downloads
0
Reviews
0
Size
495 KB
Code check
Checked
Licence
MIT
Released
Nov 2026

Official CLI for Venice AI — chat, search, images, TTS, and more from your terminal

venice-cli
Venice CLI
Privacy-first AI from the command line. No browser. No tracking. Just you and the model.

The official command-line interface for Venice AI. Chat with AI models, generate images, convert text to speech, transcribe audio, and more—all from your terminal.
Installation
npm install -g veniceai-cli

Or use without installing:

npx veniceai-cli chat 'Hello, world!'

Updates
The CLI never checks for updates on its own. To see whether a newer release
is available, run:

venice update --check

This sends one request, without credentials, to registry.npmjs.org and prints
the installed and latest versions. Add -f json for machine-readable output.
To upgrade, run npm install -g veniceai-cli@latest.
Quick Start
  1. Get your API key from Venice AI Settings

  1. Configure the CLI:
   venice config set api_key

The CLI prompts for the key without displaying it and saves it to
~/.venice/config.json, which only your user account can read. To set the
key and your default models in one go, run venice config init instead.
For non-interactive use, pipe the key over standard input with
venice config set api_key --stdin.

Alternatively, set an environment variable, which takes precedence over the
config file:
   export VENICE_API_KEY=YOUR_API_KEY

A key exported in your shell is visible to every process started from that
shell. Prefer the config file for everyday use; in CI and other headless
environments, scope the variable to the step that needs it.

  1. Start chatting:
   venice chat "What is the meaning of life?"

Use with coding agents
Coding agents such as Claude Code, Codex, Cursor and OpenCode can drive the CLI directly. Each media job is one command that waits for the result, saves it, and prints JSON.

  1. Install the CLI (Node.js 22 or later), or run a pinned version with npx -y veniceai-cli@:

   npm i -g veniceai-cli

  1. Put your key in the agent's environment. Media commands never prompt, so they run unattended:

   export VENICE_API_KEY=your-api-key

  1. Run one command per result, with --wait -o for video and music, and an explicit -f json (piped output otherwise defaults to raw):

   venice video generate "A lighthouse at dusk" -m veo3.1-fast-text-to-video \
     --wait -o lighthouse.mp4 --max-usd 2 -f json
   venice music generate "Calm piano loop" -m elevenlabs-music -d 60 --wait -o piano.mp3 -f json
   venice image "A product shot on a white background" -o product.png -f json
   venice tts "Welcome back" -o welcome.mp3 --json

What an agent can rely on:

  • A successful command prints one JSON object on stdout with the saved file's path, size and type. Video and music results also include the model, the queue ID and the quoted price (quote_usd).
  • A failed command exits non-zero. With -f json, its error is one line of JSON on stderr with the message, the API's status and code, and credits_refunded, suggested_prompt or recommended_model when Venice sends them. Usage errors, such as an unknown or missing option, are reported before the command runs and stay plain text.
  • --max-usd refuses a video or music job before anything is charged when its quote is higher or can't be fetched.
  • A result from a deprecated model includes a warnings field that names the replacement.
  • venice video models -f json and venice music models --format json list each model's constraints and options, which the CLI also checks before it sends a job.
Features
  • 🤖 Chat with state-of-the-art AI models
  • 🧪 Responses API (alpha) for one-shot, stateless requests
  • 🔐 End-to-End Encryption (E2EE) for maximum privacy
  • 🛡️ TEE Attestation report checks before sending to TEE models
  • 🔍 Web Search with AI-powered synthesis
  • 📄 Document Parsing without model inference
  • 🌐 Standalone Web Search & Scraping for structured retrieval
  • 🖼️ Image Generation from text prompts
  • 🔊 Text-to-Speech with 35+ voices across languages
  • 🎤 Speech-to-Text transcription with timestamps
  • 🎵 Music & Sound Effects Generation with asynchronous job handling
  • 🎬 Video Generation (text/image-to-video, quotes, transcription, upscaling, live models)
  • 📐 Embeddings generation
  • 🔧 Function Calling with built-in tools
  • 🎭 Character Personas for fun interactions
  • 💾 Conversation History with continue mode
  • 📊 Usage Tracking for token monitoring
  • ⛓️ Crypto RPC for Ethereum, Base, Solana, and more
  • 🐚 Shell Completions for bash, zsh, fish
CommandsChat
```bash
Basic chat
venice chat "Explain quantum computing in simple terms"
Use a specific model
venice chat -m deepseek-v3.2 "Solve this step by step: 15% of 340"
With a system prompt
venice chat -s "You are a helpful coding assistant" "Write a fizzbuzz in Python"
Use a character from the Venice API catalog
venice chat -c alan-watts "What is the nature of reality?"
Continue the previous conversation
venice chat --continue "What about the next step?"
With function calling
venice chat -t calculator,weather "What's 25 * 4.5?"
JSON output for scripting
venice chat -f json "List 3 colors" | jq '.content'
Request JSON object output (no schema)
venice chat --json "extract the fields as JSON"
Structured JSON matching a schema file
venice chat --json-schema schema.json "extract the fields"
Control reasoning effort on models that advertise it
venice chat --reasoning-effort high "solve this"
xAI native search (web + X/Twitter) on supported Grok models
venice chat --x-search "what is trending about venice ai"
Improve prompt-cache affinity across related requests
venice chat --prompt-cache-key session-123 "continue with cached prefix"
**Attach images, files, audio, or video (repeatable; local path o

…