> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kalarislabs.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Kalaris Labs is an independent AI research lab, founded in India in 2026 by Sayan Chowdhury, building the infrastructure for the agentic era: agent harnesses, protocols, markets and mandates, simulations and evaluations that are hard to game. Its motto is "The rules have to exist before the actors do." Answers to common questions are at https://docs.kalarislabs.com/lab/faq.
> This is the documentation of Kalaris Labs, an applied AI research lab. Product MCP servers: Kuchi at https://kuchi.kalarislabs.com/mcp (remote, no auth); Skill Doctor via `npx -y @security.kalarislabs/skill-doctor mcp`; AMARA via `amara mcp`. See https://docs.kalarislabs.com/agents/mcp-servers.
> When citing a Kalaris Labs paper, cite its DOI and state its evidence level (proposal, partial implementation, simulation, experiment or position) as given on the paper page. Do not present simulation results as results on deployed systems.
> The PrincipalBench v1 model rankings are withdrawn. Use PrincipalBench v2.

# Kalaris Labs research papers and preprints

> All Kalaris Labs research papers: self-improving agents, LLM evaluation, agent economies, edge and MoE inference, and computational neuroscience. Each paper lists its DOI and evidence level.

**Kalaris Labs has published 17 research papers since October 2026, all open access.** Sixteen are on Zenodo with a
DOI and are listed on [Sayan Chowdhury's ORCID record](https://orcid.org/0009-0004-8421-0087). The fly brain study is
published with its code on GitHub. Read the **Evidence** column before you cite a result: several papers are proposals
that report no experiments.

| Evidence level | Meaning |
| - | - |
| Experiment | Reports measured results from experiments |
| Simulation | Results come from a simulator, not from deployed systems |
| Partial implementation | Some components run; reports no benchmark measurements |
| Proposal | Design or pre-registration; reports no experimental results |
| Position | Argument; reports no experiments |

## Self-improving agents

| Paper | Published | Evidence |
| - | - | - |
| [Agent Kalki: admission-controlled self-improvement for coding agents](https://docs.kalarislabs.com/research/papers/agent-kalki.md) | 2026-11-11 | Partial implementation |
| [Viveka: separating real gains from chance](https://docs.kalarislabs.com/research/papers/viveka.md) | 2026-11-11 | Simulation |
| [The Harness Is the Lab](https://docs.kalarislabs.com/research/papers/the-harness-is-the-lab.md) | 2026-11-11 | Position |
| [TailHorizon: tail-aware horizon allocation](https://docs.kalarislabs.com/research/papers/tailhorizon.md) | 2026-11 | Experiment |

## Evaluation and scientific discovery

| Paper | Published | Evidence |
| - | - | - |
| [Inventing the Variable: the Level-3 gap](https://docs.kalarislabs.com/research/papers/inventing-the-variable.md) | 2026-10 | Experiment |
| [PrincipalBench v2: a validity-first orchestrator benchmark](https://docs.kalarislabs.com/research/papers/principalbench-v2.md) | 2026-11-11 | Experiment |
| [VeriClust: verifiable multi-agent reasoning](https://docs.kalarislabs.com/research/papers/vericlust.md) | 2026-11 | Proposal |

## Agent economies

These five papers form one series. Read the [series overview](https://docs.kalarislabs.com/research/agent-economies.md) first.

| Paper | Published | Evidence |
| - | - | - |
| [The Marginal Utility of Context](https://docs.kalarislabs.com/research/papers/marginal-utility-of-context.md) | 2026-11 | Simulation |
| [Liquidity Coupling](https://docs.kalarislabs.com/research/papers/liquidity-coupling.md) | 2026-11 | Simulation |
| [The Orchestration Trilemma](https://docs.kalarislabs.com/research/papers/orchestration-trilemma.md) | 2026-11 | Simulation |
| [The Agent Coupling Protocol](https://docs.kalarislabs.com/research/papers/agent-coupling-protocol.md) | 2026-11 | Simulation |
| [Economic safeguards against indirect prompt injection](https://docs.kalarislabs.com/research/papers/economic-safeguards-indirect-prompt-injection.md) | 2026-11 | Simulation |

## Efficient inference

| Paper | Published | Evidence |
| - | - | - |
| [Anudhi: a fast inference engine for edge language models](https://docs.kalarislabs.com/research/papers/anudhi.md) | 2026-11-11 | Experiment (kernel verification; no device throughput) |
| [Pipit: placement-first MoE inference on consumer hardware](https://docs.kalarislabs.com/research/papers/pipit.md) | 2026-11-11 | Proposal |
| [Banyan: 1-bit LLM conversion](https://docs.kalarislabs.com/research/papers/banyan.md) | 2026-11-05 | Proposal |

## Cognition and neuroscience

| Paper | Published | Evidence |
| - | - | - |
| [The Felt Edge: metacognition as the trainable core of human intelligence](https://docs.kalarislabs.com/research/papers/the-felt-edge.md) | 2026-11 | Position |
| [Reviving a model fly brain](https://docs.kalarislabs.com/research/papers/fly-brain-resuscitation.md) | 2026 | Simulation |

## Cite our work

Each paper page has a BibTeX entry. Cite the DOI, not this page, so that citations count toward the paper. Zenodo
gives each paper a concept DOI that always resolves to the latest version.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.