Skip to main content
Short answers to the questions people and agents ask most about Kalaris Labs. Each answer links to the page with the full detail.

About the lab

What is Kalaris Labs?

Kalaris Labs is an independent AI research lab, founded in India in 2026, building the infrastructure for the agentic era. Its research covers agent harnesses, continual and reinforcement learning, multi-agent simulation, game theory and the economics of agents, agent protocols, evaluations, and AI safety and security. See about the lab.
The period in which AI agents act rather than only answer: writing and shipping code, transacting, negotiating, and coordinating with other agents on behalf of people and institutions. Those agents need infrastructure the current internet was not built for, including shared protocols, enforceable mandates and evaluations they cannot game.
The foundations agents need before they can be trusted with real work: harnesses they run in, protocols they use to identify themselves and prove what they did, markets and mandates that set what they may spend, and evaluations they cannot game. Every earlier era of computing got this layer after the damage. This one has to be built before.
“The rules have to exist before the actors do.” When software acts at machine speed, the rules, protocols and checks agents operate under have to be in place before agents are deployed, not added after something breaks. Read the manifesto.
Sayan Chowdhury, an Indian agentic AI researcher and systems architect who directs the lab’s research on agentic systems architecture, execution harnesses and the economic layer of autonomous agent systems. He is the author of the lab’s preprints. See Sayan Chowdhury’s profile for his LinkedIn, X, ORCID, GitHub and Hugging Face.
India. The lab builds in the open from India because it believes the foundations of the agentic era should be a public good, the way India’s payments network is one that any bank or app can build on.
Agent protocols, game theory, the economics of agents, and AI security and safety. The flagship project, Agents on Web, is a planned public simulation where agents live in a shared world and anyone can watch where they break; work begins in 2027. New results appear in the changelog.
Email hello@kalarislabs.com for research collaboration, press or partnerships. Report security issues through each repository’s SECURITY.md, not in public issues.

Research

Sixteen papers are open access on Zenodo with a DOI; the fly brain study is published with its code on GitHub. Every paper is listed on the research page.
Cite its DOI and state its evidence level (proposal, partial implementation, simulation, experiment or position), both shown on the paper’s page. Do not present simulation results as results on deployed systems.
No. The PrincipalBench v1 model rankings are withdrawn. Use PrincipalBench v2.

Tools and agents

Yes. All code is public at github.com/KalarisLabs. Licenses differ by repository; see open source.
Run npx skills add KalarisLabs/research-agent-skills. There is no research-agent-skills package on npm. See installation.
Run npx @security.kalarislabs/skill-doctor scan ./my-skill. Skill Doctor runs offline by default and does not use a model to decide what is safe. See the Skill Doctor quickstart.
Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and Windsurf.
Connect the docs MCP server at https://docs.kalarislabs.com/mcp, read llms.txt or llms-full.txt, or append .md to any page URL. See agent access.
Last modified on November 15, 2026