Research Papers Prime Agent is an open harness for long-horizon coding agents built on the Recursive Language Model idea: a persistent IPython session does test-time computation, while histories, memories, skills and subagent specs carry over between runs.
FreeToken rethinks local inference around two realities of edge AI: agent workloads keep changing shape, and every machine's resource balance differs. It co-designs model layout, expert residency, CPU-GPU execution and memory management instead of committing to a fixed offloading strategy.
ABot-World-0 is an action-conditioned video world model trained on multi-source data from AAA games, simulation engines and internet videos, controlled through raw keyboard input with reference-character memory for consistent third-person rollouts. The authors report streaming 720p at up to 16 FPS on one RTX 5090 with 1.2s action-to-first-frame latency, targeting real-time long-horizon closed-loop interaction.
Ouroboros is a self-developing coding-agent harness whose tools, prompts, context assembly and even core implementation improve through reviewed commits that become its later runtime, with two evolution modes: recursive free evolution and experience-driven core evolution. The authors also report a 161-day live deployment across seven human communication surfaces, arguing safety guardrails must stay authoritative while the agent rewrites its own code.
This harness maintains an explicit task state outside the agent's execution context and updates it only with facts independently verified from the environment, using a Manage-Execute-Audit loop so mistaken self-assessments do not snowball across many steps.
After Meta's closed Muse demo showed what a single-person agent with lasting memory could look like, this report reconstructs how Muse is built from public sources, then ships nanoMuse as its GPL-3.0 open counterpart: one agent across the devices you own, sharing a single conversation over a relay anyone can self-host. Every action passes a Sentinel gate before it executes, memory is stored as plain files the owner can read, and the underlying model is the user's choice.
Making a long film with a video-generation agent breaks down at the seams: passing prompts or a few recent frames between shots gives the model neither a shared spatial frame for revisiting a scene nor a time-indexed record for replaying an earlier event. ATAS compiles a screenplay into shot requests with explicit location, camera, and event-time references, and keeps two memories behind the generation loop: a scene memory of registered static geometry and an event memory of dynamic observations indexed by scene, event, and time. Each generated clip is reconstructed and registered back, admitted only after review and state checks. The authors report that this improves all measured cross-shot consistency metrics across four video generators, outperforms compared agents and 3D-guided methods on spatial revisits and event recapture, and earns the highest human ratings on all six evaluated criteria.
AutoDesign treats multimodal conference-poster generation as a long-horizon agentic process: a meta-harness optimizer reads rollout feedback and rewrites the harness the code agent runs on. On the authors' 100-paper PosterBench it scores above Claude Design, the learned harness lifts seven different code-agent configurations, and a fully autonomous loop makes 253 tool calls in 40 minutes for under $3.
Raven argues that hand-crafted agent harnesses are too domain-coupled to scale with long-horizon, cross-domain workloads. It treats each model–harness pair as a composable unit of intelligence: a Host Agent decomposes goals and coordinates specialized sub-agents, a host archive and EverOS retain experience across tasks, and Skill Forge turns that experience into reusable procedures — with theory on expanding reliable task coverage under one resource budget.
This 66-page position-style paper formalizes harness self-evolution: an agent keeps its model weights fixed but continually revises its own software scaffolding, including prompts, memory, workflows, executable components, and plugins. Accepted revisions persist across runs, and the authors separate what can be updated from the proposal and admission mechanisms, analyzing reliability with PAC-learning-style bounds and an admission rule that requires fresh independent validation data. The paper argues this demands continual validation plus independent safeguards on runtime privileges and consequential actions.
EvoOntology tackles the agent-data gap: agents can see column names and file paths through generic tools, but not what the data actually means. It wraps an ontology (schema, content, and tool layers) as an MCP server the agent can query at runtime, uses a builder agent to construct it autonomously, and keeps refining it with attribution-guided edits that are only accepted after a paired evaluation against the backbone. The authors report consistent gains over strong baselines and hand-built semantic layers across three data-agent benchmarks and four LLM backbones.
WeMM-Embedding is a family of universal multimodal embedding models (2B, 4B and 9B) covering text, images, video, visual documents and arbitrarily interleaved inputs, with flexible output dimensions, trained by large-scale alignment followed by refinement with curated relevance supervision.
OpenGCUA rebuilds the computer-use agent stack in the open, motivated by the observation that frontier systems like GPT-6 Astra's computer use are closed. Its core idea is to let the model choose between visual GUI actions and direct code execution, so intermediate results can steer later steps; it also proposes harness-to-model decision transfer, moving selected observation and task-state decisions out of fixed harness rules and into the model itself. The authors report that on OSWorld-v2 their full configuration improves task scores across all eight evaluated models (including GPT-6 Astra and Claude Opus 5), raising mean partial score by +19.6 points and cutting reported steps per task by 53.0% on average versus batched GUI-only execution. Alongside the implementation and prompts they release evaluation trajectories from all eight models, plus early versions of a real-device harness and a cross-device SDK covering computers, mobiles, and browsers; on four real macOS tasks their harness shows selected Qwen3.8-Max runs use 19.4 times as many reported reasoning tokens as GPT-6 Astra, which they highlight as a gap in real-device execution efficiency.
Apodex 1.1 targets what the authors call working capability: it scales up the diversity and verifiability of executable file, search and code environments, while training agents to decompose long-horizon tasks, delegate parallel work and integrate asynchronous results.
FDE-Gym tests agents on the last mile of AI delivery instead of well-specified issue tickets. It builds 95 end-to-end delivery tasks from 20 customer scenarios across 18 sectors, each adapted by industry experts from real delivery projects; agents interact with scripted stakeholders and modify production-like customer systems, and a delivery is scored by how much it improves the customer's own success metric, normalized between 0 (unchanged) and 1 (reference solution), with penalties for excessive messages, trial runs, and waiting. The authors report the best of 13 model configurations from seven providers scores only 0.56 on average and fully resolves just 14 of 95 tasks, with unstated-constraint discovery as the main bottleneck; even on the 23 tasks where the best configuration extracted every key fact, it reached the 0.8 uplift threshold on only 10.