- Venice's default chat is now an agent that breaks complex tasks into steps, selects the right tool for each step, and delivers results across text, images, video, search, and more in a single conversation.
- Multiple agent models available (Kimi K2.5 free/private by default).
- Image, video, web search, and file analysis tools activate automatically. For image and video generation, the system selects from multiple specialized models based on the request.
- Classic chat remains accessible.
The Venice default chat is now agent-powered.
Instead of a single model responding to a single message, the agent breaks tasks into steps, uses the right tool for each step, and produces results across text, images, video, and more in a single prompt.
Try it now at venice.ai/chat/agent
What is agentic chat
When you send a message, the agent receives a config of available tools and decides how to handle your request. The agent then runs a loop: reason about the task, pick a tool, execute it, observe the result, and decide what to do next. It might search the web, generate an image from what it found, then animate that image into a video. Each step streams to you in real-time as it happens. This repeats until the agent decides it no longer needs to call any more tools.
When the agent decides to use a tool, it calls "generate an image" or "search the web," and the agent selects from the models available to your account. Your subscription tier and settings determine which models are in the pool. You can also tell the agent which model you want it to use (more information about the models in the next section).
For image and video generation, the agent selects the best model for the request from the available pool. You confirm video generation before it starts (since it costs credits). You can keep chatting while a video renders, but not while an image generates.
Models under the hood
When you select an agent model, that model powers the entire conversation and decides which tools and sub-models to invoke. You pick one model at the start; everything else activates automatically based on what the task needs.
The model dropdown lets you choose which model powers the agent:
Kimi K2.5. Moonshot AI. 1T total parameters, 32B active (MoE). 256K context. Private inference, zero data retention. Free for all users. Default.
Claude Sonnet 4.6. Anthropic. Balanced performance across coding, reasoning, and general tasks. 1M context. Anonymized inference. Costs credits.
Claude Opus 4.7. Anthropic. Most capable model for complex multi-step reasoning and agentic coding. 1M context. Anonymized inference. Costs credits.
Free vs. premium models
Some models in the agentic workflow are free, while others require credits. The agent defaults to free models whenever possible. If a task would benefit from a premium model, you are prompted to enable the premium model.
A banner appears asking "Enable Premium Image Models?" with the specific premium model names listed. You click "Enable" to opt in, or dismiss to stay on free models only.
Once enabled, the preference persists across sessions (you can disable it at any time in Settings > Usage).
Image generation
The agent picks the right image model based on your request:
Image editing
A dedicated upscaler model handles "make this higher resolution" requests. Up to 4x. Free for all users.
Video generation
All video generation costs credits. The agent will request video generation regardless of your balance, but you need sufficient credits to complete the render.
All three support text-to-video and image-to-video.
Multimodal workflows with agentic chat
The agent has access to every creative and analytical tool on Venice.
Ask it to generate images, edit them, upscale them, and turn them into a video. Search the web and scrape pages. Parse uploaded files, pull crypto market data, create Mermaid diagrams, and more.
Supported file uploads: PNG, JPEG, WebP, GIF, PDF, DOCX, XLSX, CSV, JSON, XML, YAML, Markdown, plain text. 25 MB per file, 50 MB total per message.
The interesting part is chaining modalities within a single conversation, and even within a single prompt.
"Research the latest trends in product photography, generate 3 images showcasing these trends, then animate the images into 10-second videos with slow camera movement."
The agent researches, thinks, generates three images, and kicks off three videos, all from a single prompt.
How agentic chat handles privacy
Venice's privacy model is built into the agentic chat infrastructure. The system is designed so that even Venice itself cannot access your conversation content.
When you use the default model (Kimi K2.5), your entire conversation runs privately on Venice-controlled GPU infrastructure with zero data retention. Nothing is logged. The model processes your request, returns the response, and stores the state on your device. Venice cannot access, review, or audit these conversations because the data does not physically persist on the server side.
Conversation history is stored exclusively on the client side, in your browser. Venice's servers are stateless between requests. There is no centralized conversation database. File uploads (PDFs, images, documents) are processed in memory during your session and stored in your browser.
If you select Claude Sonnet 4.6 or Claude Opus 4.7, Venice operates as an anonymizing proxy: all identifying metadata is stripped before your request reaches Anthropic. Anthropic sees a request from Venice's infrastructure with a non-reversible hash, not from you. Note: if you switch your agent from Kimi K2.5 to Claude mid-conversation, Anthropic does receive your full conversation history from that session.
When the agent calls an external tool (Nano Banana Pro for images, Seedance V2 or Wan 2.7 for video), the external provider only receives the tool call parameters: the prompt, resolution, and generation settings. It does not receive your conversation history, your identity, or any other context. All free image models (Z-Image Turbo, Chroma, Wai Illu), the upscaler, run on Venice infrastructure with the same zero-retention guarantee. Grok Imagine video is also fully private.
What about classic chat?
Classic chat is not going away. Click "Use Classic hat" in the interface to switch back at any time.
Classic chat still offers:
- The full 230+ model catalog
- Direct single-model selection for when you want a specific model for a specific task
- Reference-to-video (R2V) and specialized video generation modes not yet available in agentic chat
- Characters and character modes
- All image editing tools (inpaint, multi-edit, upscale, background removal)
- Music and audio generation
Use agentic chat for diverse, general-purpose, or complex work with many steps.
Use classic chat when you want granular control over which specific model handles your request.
And Venice Studio is the place to be for AI filmmaking workflows, giving you similar granular control over models and parameters.
Agentic chat is now live
Agentic chat is now the default for all new users and rolling out to existing users over the coming days. If you don't see it yet, you will soon. No action needed on your end.
Try it: venice.ai/chat/agent
