Weekly Trending

Prime Agent is an open-source coding and research agent built around recursive language models, a persistent REPL, durable harness state, and background subagents.

cloudflare/computer is a TypeScript runtime that gives an AI agent a provisioned computer surface.

An open model stack for long-form speech recognition and low-latency synthetic voice, surfaced as a signal for modular and governable voice-agent infrastructure.

ego-lite is a local browser that gives humans and AI agents parallel workspaces and exposes browser actions to coding agents. Its design makes session data, cookies, login state, and human oversight explicit parts of the operating surface.

Hallmark is an anti-AI-slop design skill for Claude Code, Cursor, and Codex. It encodes visual quality-control constraints as a machine-readable skill layer.

Uber's open ADR stack turns agent intent, tool calls, execution traces, and attack tests into an inspectable governance layer.

RAGFlow is an open-source RAG engine that combines retrieval-augmented generation with agent capabilities to serve as a context layer for LLM applications. It is relevant to knowledge-base products, research ingestion, and self-maintaining wiki systems.

Semantica is graph-native infrastructure for accountable AI systems, combining context graphs, deterministic reasoning, provenance, decision records, and MCP access.

WrenAI is an open-source governed text-to-SQL and generative BI layer for agents across warehouse and database sources. Its semantic context layer is intended to make business-data questions more trustworthy and controllable.

NatureBench evaluates whether coding agents can reproduce or improve work from 90 tasks distilled from Nature-family papers. It finds that agents often translate unfamiliar science into familiar supervised-prediction shapes instead of devising new methods.

A paper proposing claim-level provenance for multi-agent knowledge systems, grading transmission links, corroboration, and uncertainty rather than treating a citation list as sufficient evidence.

AGORA is a benchmark for agentic workplace document reasoning across 9,664 authentic documents and 372 million tokens. It tests deliberate exploration of messy corporate archives rather than brute-force use of a single context window.

AOHP is an open-source OS-level agent harness that treats agents as first-class operating-system actors rather than app-bound guests.

A paper proposing a long-horizon mobile GUI agent that treats proactive context management as actions chosen by the same policy that performs UI work.

The paper proposes an execute-distill-verify approach to agentic experience learning, separating observed execution from distilled memory and independent verification.