Execution, not just answers
Use Plan, Goal, and Ultra modes when the work needs alignment, completion criteria, or aggressive multi-agent delegation.
Open-source desktop workspace · Created by AIVAX Labs
Keep the work around the model in view.
AVI keeps the conversation beside the project, files, and changes it is working on—then gives tools and agents a visible place to operate.
Explore AVI on GitHub
MIT licensed Built in the open
- Conversation
- Project
- Files
- Diff
- Branch
Why AVI
The model is only one part of the work.
A conversation is only one part of serious AI work. AVI keeps the surrounding system close: the project, its instructions, the tools being called, the tasks in motion, and the agents doing the work.
That makes AVI less like another chat client and more like a visible operating layer for serious AI work.
Context that arrives ready
AVI discovers project instructions, skills, workflows, AGENTS.md, and MEMORY.md around the work.
Orchestration you can inspect
Follow tool calls, tasks, parallel threads, and sub-agent progress instead of receiving an unexplained final answer.
The workspace around the model
See the work behind the answer.
One conversation can carry the objective, the tools used to investigate it, the tasks still open, and the agents working in parallel.
Tools, tasks, sub-agents, model, permissions, project path, and branch stay attached to the active conversation.
Choose how the work runs
Different problems deserve different operating modes.
Research, compare approaches, and align on an execution path before implementation begins.
Define completion criteria and keep the work moving until the condition is met.
Delegate across focused sub-agents, challenge the result, refine it, and validate again.
Everything the conversation needs
A complete workspace with provider choice.
Connect the providers you choose. Give each project its own context and tools. Keep consequential actions behind the permission level you set.
Multiple providers
Connect a ChatGPT subscription or services implementing OpenAI-compatible Responses and Chat Completions APIs.
Project-scoped MCP
Configure servers globally or by folder, inspect their tools, and diagnose slow or failed startup.
Communicating sub-agents
Choose model and reasoning levels for delegated work while agents exchange findings with the orchestrator and each other.
Discovered context
Make local and global skills, workflows, instructions, and memory available without rebuilding every prompt.
Permission controls
Choose when consequential tools should ask, be allowed for the current action, or receive full access.
Quick chats and side chats
Ask a fast standalone question or fork a side conversation without interrupting the main thread.
Files, Git, tasks, usage
Keep the repository state and operational detail visible in the side panel while the conversation continues.
Local conversation state
Conversation state stays on your machine; model requests go only to the providers you configure.
AVI · Created by AIVAX Labs
Open by design
Built in the open. Yours to inspect.
AVI is an MIT-licensed project created by AIVAX Labs. Read the source, follow its development, or shape the workspace around the way you work.
View aivaxlabs/avi