Sentient Labs logo

Sentient Labs

Sentient Labs is an active open-source AI reasoning lab building research, developer tooling, agent-evaluation products, and AI-security primitives.
Singapore, Singapore

Eldros


Project Links


Description

Eldros provides closed-loop reliability tooling for AI agents, including scenario simulation, workflow-level evaluation, production monitoring, failure-pattern analysis, regression testing, and targeted improvement recommendations for regulated and customer-facing workflows.

Category: AI Agent Platform

Sentient Arena


Project Links


Description

Sentient Arena crowdsources AI reasoning research through standardized challenges. Builders create skill packages—including instructions, scripts, sub-routines, and verification logic—to improve coding agents on difficult enterprise and real-world reasoning tasks. Submissions are evaluated under shared conditions and compared through challenge leaderboards.

Category: AI Agent Platform

ROMA (Recursive Open Meta-Agent)


Project Links


Description

ROMA represents work as dependency-aware recursive task trees. Its modular Atomizer, Planner, Executor, Aggregator, and optional Verifier components separate orchestration from model and tool selection, enable parallel execution and traceable context flow, and support self-hosted use through a Python package, CLI, REST API, and optional persistence and observability services.

Category: AI Agent Framework

Sentient Chat


Project Links


Description

A user-facing multi-agent AI application that unifies AI chat with workflow “Spaces,” drawing on GRID’s network of agents, models, data sources, and tools.

Category: AI Agent Platform

CryptoAnalystBench


Project Links


Description

CryptoAnalystBench evaluates AI-generated crypto and DeFi analysis with an LLM judge across relevance, temporal relevance, depth, and data consistency. It contains 198 production queries across 11 categories, an agentic evaluation harness, citation and hallucination checks, annotated queries, judge rubrics, and a seven-part error taxonomy.

Category: Developer Tooling

ROMA


Project Links


Description

ROMA recursively decomposes complex goals into dependency-aware subtasks, executes them through configurable agents or tools, and aggregates results through a transparent hierarchical workflow. Its open-source ROMA-DSPy implementation provides modular agent components, configuration profiles, toolkits, CLI, REST API, and optional production services.

Category: AI Agent Framework

EvoSkill


Project Links


Description

EvoSkill runs coding agents on benchmark tasks, analyzes failed trajectories, proposes skill or prompt changes, evaluates candidates on held-out data, and retains stronger agent configurations on a Git-tracked frontier. It supports Claude Code, Codex CLI, OpenCode, OpenHands, Goose, Harbor, and other compatible agent runtimes.

Category: AI Agent Framework