Chats
A chat workbench. Run a prompt against one model or many side by side (OpenAI, Gemini, Mistral, Ollama, HuggingFace), save and reuse prompts, and optionally bind an agent graph to a chat so it runs on every turn.
Chatty the Lab is a multi-agent AI research platform. It brings everything you need to design, run, and study agentic systems into one place: a chat workbench for comparing models, a visual editor for building agent graphs, a research toolkit for experiments and labeling, and the settings that tie your providers and workspaces together.
The platform is organized around four pillars. Each pillar is a self-contained area of the app, but they share the same data and workspaces, so an agent you design in the Studio can run inside a chat or power an experiment without being rebuilt.
Chats
A chat workbench. Run a prompt against one model or many side by side (OpenAI, Gemini, Mistral, Ollama, HuggingFace), save and reuse prompts, and optionally bind an agent graph to a chat so it runs on every turn.
Studio
A visual agent-graph editor. Drag nodes onto a canvas, connect them, configure each one, then run the graph and watch live execution — with real-time multiplayer collaboration.
Research
The research toolkit: typed datasets, parameter-sweep experiments and trials, human labeling jobs, and a collaborative logbook for writing up your work.
Settings
Your account, provider API keys, model pricing catalog, and local Ollama / embedding setup — plus workspace and team management.
Dive into a pillar:
Every resource in Chatty — chats, agents, graphs, datasets, experiments, and even some settings — is scoped to a workspace. A workspace is either:
A WorkspaceSwitcher in the navbar lets you switch the active workspace at any time. When you switch, the entire app re-scopes: you only see the agents, datasets, and resources that belong to the workspace you are currently in. This keeps personal experiments separate from shared team work.