Chat with many models
Run prompts against OpenAI, Gemini, Mistral, Ollama, and HuggingFace side by side, or bind an agent graph to a chat for richer behaviour.
Chatty the Lab
Chatty the Lab is a research-oriented platform for building and studying multi-agent AI systems. Built at CiTIUS (University of Santiago de Compostela), it pairs a visual agent-graph studio with a full research toolkit: experiments and parameter sweeps, typed datasets, human labeling jobs, and a collaborative logbook — all scoped to personal or team workspaces.
What you can do
Section titled “What you can do”Build agent graphs
A visual canvas (the Studio) with 26 node types and 6 edge types: LLMs, tools, conditions, loops, panels, orchestrators, and more — with real-time collaboration.
Coordinate multi-agent panels
Convene round-table debates (Panel) or hierarchical delegation (Orchestrator) over a roster of agents, each with its own memory and communication policy.
Run experiments
Sweep parameters (grid / random / TPE) over a graph and a dataset, score trials, plot results, and append outputs back to datasets.
Label & annotate data
Typed datasets with CSV/JSON import, plus labeling jobs with roles, custom questions, assignments, and inter-annotator agreement.
Drive it from Python
pip install chatty-lab — declare a graph with an operator DSL, name every version of
it by its content, and push the history to a workspace.
Collaborate in teams
Personal and team workspaces scope every resource. Invite members, assign roles, and keep a shared research logbook.
Start here
Section titled “Start here” Introduction The four pillars and how the platform fits together.
Agent Studio Build visual agent workflows on the canvas.
Node reference Every node type, grouped and explained.
Research Experiments, datasets, labeling, and the logbook.
Python API Define investigations in code, version them, and send them to the platform.
Where it runs
Section titled “Where it runs”chatty-lab is a hosted platform. The public instance runs at chatty-lab.com; the CiTIUS internal one at tec.citius.usc.es/chatty. You sign in and work in a workspace — there is nothing to install.
What each workspace brings is its own credentials: the platform holds no keys of its own, so the models you can reach are the ones you have configured. See Providers & API keys.
If you want to know how it is put together rather than how to use it, Architecture is the map.