Software Engineering

DeepSeek Harness v0.1 is an MIT-licensed agent harness where every capability is a Cordis plugin. Four runtime modes, append-only session logs, and provider-agnostic model routing.

Z.ai released GLM-5.3 on August 14, 2026. The model reuses the 743B GLM-5.2 base unchanged. Every reported gain comes from scaled post-training: more long-horizon task environments, more environment types, longer training. Terminal-Bench 3.0 moves from 4.6 to 28.3, and DeepSWE v1.1 from 46.2 to 66.9. Cybersecurity moved further than Z.ai says it planned, with CyberGym at 84.5% and ExploitBench more than doubling to 54.4%. Weights arrive in about two weeks.

Cactus Compute released Needle 2, an open 45M-parameter model for tool calling, device use, and structured extraction. The full model is a single 14MB binary that runs a session in about 28MB of RAM. It leads both Seal-Tools splits while targeting hardware with no GPU and no NPU.

Google has released Gemini 3.7 Flash, a refinement of Gemini 3.6 Flash with algorithmic improvements to its reasoning core. It handles text, images, audio, and video across a 1M-token context window with 64K-token output, and supports customizable thinking configurations. Coding results move notably: 43.6% on FrontierCode 1.1 Main versus 34.4%, 65.3% on DeepSWE v1.1, and 1588 Elo on WebDev Arena. Document and workflow evals gain more — GDP.pdf goes from 22.0% to 34.0%, AutomationBench from 17.0% to 30.4%. GPT-5.6 Terra still leads on terminal and computer-use agents. The model is API and enterprise only, with no open weights, at an introductory $0.75/$3.75 per 1M tokens through December 31, 2026.

NVIDIA's open 30B MoE targets the agent execution layer, with Switchyard routing each step to the cheapest capable model.

webAI has released TwIL-LM, a family of formal-logic models at 1.7B and 3B parameters that translate English into first-order logic and check whether conclusions follow from premises. The 3B runs on CPU or 4GB of VRAM; the 1.7B downloads at 1.06GB. Both ship under a non-commercial license. The model card also shows that the headline benchmark scores belong to an unreleased checkpoint, not the weights on the Hub.

Mistral AI has released Shieldstral 1.0 3B, an open-weights, policy-adaptive multimodal safety classifier that frames content moderation as a single yes/no question instead of a fixed harm taxonomy. Operators supply the policy as a plain-language query at inference time and get back a calibrated safety score from one forward pass — no retraining required to re-target the model. Built on Ministral-3-3B-Base-2512 with a Pixtral vision encoder and trained on roughly 54.1M samples, it reports 84.9% average F1 on text safety (matching GPT-OSS-Safeguard-20B), 83.8% on multimodal safety, and 91.3% on Mistral's adaptability benchmark — while fitting in 16GB of VRAM under an Apache 2.0 license.

Tencent Cloud has open-sourced TencentDB Agent Memory v2.0, a team-level memory hub that turns conversations, documents and code into four governed, reusable assets — Chat Memory, Skill, LLM-Wiki and Code-Graph. It is MIT-licensed, self-hosted via Docker, and integrates with Claude Code, OpenClaw, Hermes and CodeBuddy. The differentiator is not retrieval but governance: ACL-based visibility decides which agent gets which asset, and which version is valid.

NVIDIA Labs has open-sourced NOOA (NVIDIA Object-Oriented Agents), a model-agnostic Python framework for building AI agents. Agent development today is split across prompt templates,...

Microsoft has open sourced code-testing-generator, a polyglot unit-test agent shipping in the MIT-licensed dotnet/skills repository. It reads a repository before writing anything — detecting the language, test framework, existing conventions, and the real build and test commands — then plans, writes, runs and validates the tests it produces. On Microsoft's internal 152-task benchmark it completed 140 tasks against 120 for stock GitHub Copilot on the same model, with the gain concentrated almost entirely in vague prompts and diff-targeted requests.

Cloudflare has released Kitesurf, a stateless web browser built specifically for AI agents that runs entirely in V8 isolates on Cloudflare Workers, with no Chromium underneath. The browser drops human-facing features like tabs and extensions in favor of what agents need: machine-readable content, scalability, and isolation. Built in 12 weeks using Rust components like Blitz, Stylo, and Boa JS, it already passes 215,000+ Web Platform Tests. Benchmarks show 3.1–3.8× less CPU and 4.7–7.0× less memory than Chromium on screenshots and HTML extraction. Existing Puppeteer, Playwright, and MCP clients work by adding a single browser=kitesurf parameter, free while in beta.

Prime Intellect has open-sourced Prime Agent, a coding and research harness built on two abstractions: the Recursive Language Model, which turns sub-agent calls into functions inside a persistent IPython kernel, and the Continual Harness, which lets the agent edit its own prompts, skills, memory, and sub-agent specs mid-run. With Opus 5 it reports 95.5% RHAE Best@1 on ARC-AGI-3, above the reported human expert baseline of 95.4%.

Most coverage of Microsoft's SkillOpt centers on its 52/52 result. The more consequential finding is in Section 4.3: the exported best_skill.md keeps working in environments it was never trained on. A Codex-trained SpreadsheetBench skill lifted Claude Code from 22.1 to 81.8, slightly above the 80.4 that harness reached training its own skill. Retention varies sharply by task type — 102% on spreadsheets, 10% on math — which is what makes the result worth reading closely.

Meta Superintelligence Labs has released Muse Code, a terminal coding agent in beta, powered by the new Muse Spark 1.2 model. Muse Code plans changes, writes code, and validates results across large repositories. Async background agents stay active for the whole session instead of spawning per task. A local append-only event log makes the runtime replay-exact and restart-safe after a crash. Muse Spark 1.2 was co-trained with the harness and trained on long-horizon, repository-scale work.

CopilotKit has published the Channels SDK, an MIT licensed library that runs an existing AG-UI agent inside Slack and Microsoft Teams. Version 0.5.0 ships five platform adapters and a documented runtime contract. This breakdown covers the verified deployment paths, the baseline requirements, and the one dependency that is easy to miss

Cursor Research has open-sourced Mixture-of-Kittens (MoK), the MoE training megakernel behind its Composer models. MoK fuses all mixture-of-experts communication and computation into a single deterministic kernel, and runs up to 2.37x faster than the strongest public baseline on GB300 NVL72 racks. It requires Blackwell SM100 or SM103 GPUs, which puts it out of reach for anyone without NVL72 capacity.

Reflex has released XY, an Apache-2.0 Python charting library that moves rendering work into a native Rust core and a WebGL2 client. It holds roughly 0.08 seconds render time from 10,000 to 100 million points, exports a 10-million-point interactive scatter at 258 KiB, and keeps exact f64 columns in Python so hover, selection, and zoom drilldown still return original rows. The library is early alpha at version 0.0.1.

Learn how to build an end-to-end security assessment pipeline for AI agent skills using NVIDIA SkillSpector and LangGraph. In this tutorial, we construct a synthetic skill marketplace, scan for malicious prompt injection, credential access, and risky dependencies, and implement custom YARA rules, baseline suppressions, and CI deployment gates.

Y Combinator has open-sourced QM, the multiplayer agent harness it uses internally across accounting, legal, events, and engineering. Released July 31, 2026 under an MIT license, QM gives each employee an isolated workspace and each Slack room its own scoped memory, files, keychain view, permissions, crons, web apps, and durable sandbox. Pi, OpenCode, Codex, and Claude Code all drive the same headless core, so deployments avoid vendor lock-in.

AI agents, MCP servers, and LLM apps break the core AppSec assumption that applications do what their code says. This guide walks through a practical see-fix-protect framework: a five-layer agentic AI attack surface map, a 12-point misconfiguration checklist, an evidence-based triage matrix, runtime guardrails, and system prompt hardening — plus a maturity self-assessment aligned to NIST AI RMF, OWASP AIMA, ISO/IEC 42001, and the EU AI Act.

JetBrains Research has open-sourced KotlinLLM under the Apache License 2.0. The IntelliJ IDEA plugin prototype adds Smart macros, asLlm and mockLlm, whose bodies are generated Kotlin source rather than live model calls. The plugin captures runtime values through JDI, asks an LLM agent for a narrow code update, compiles it, and redefines the loaded class. Covered scenarios then run as plain Kotlin with no further inference call. On an adapted Spring Petclinic project, 24 of 24 scenarios completed with a 100% hot-reload success rate and roughly 1% runtime overhead.

Nous Research has released Hermes Agent support for Buzz, Block's open source, self-hostable Nostr workspace where humans and AI agents share the same channels. Three integration paths cover Desktop runtime, relay bridge, and a native gateway platform that preserves Hermes memory, skills, approvals, and cron delivery.

Tencent has released AngelSpec, an open-source torch-native framework for training speculative-decoding draft models across six architectures. It introduces DFly, a block-diffusion drafter with hybrid target conditioning and a hidden-correction autoregressive head, and integrates D-cut for runtime-adaptive verification budgeting. On HY3-295B-A21B with TP=8, DFly-8 delivers a 1.98–2.40× speedup over autoregressive decoding across concurrency 4 to 64.

Marktechpost AI has released Token Saver, an open-source MCP extension for Claude Desktop that uses local Hybrid RAG to slash PDF token consumption by up to 99% while ensuring absolute document privacy.

Three terms now compete for the same line in AI engineering job descriptions. Prompt engineering is the established one. Loop engineering entered the AI...

Moonshot AI has open-sourced MoonEP, an Expert Parallelism (EP) communication library for distributed Mixture-of-Experts (MoE) workloads. The team announced the release as a library...

Liquid AI released two open-weight bidirectional encoders, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M. Both carry an 8,192-token context and are built on the LFM2 hybrid backbone. The 350M ranks fourth of 14 models on a 17-task GLUE, SuperGLUE, and multilingual suite, behind only larger models. The 230M clears one 8K-token forward pass on CPU in about 28 seconds.

In this tutorial, we configure and operate Kimi CLI as a fully non-interactive AI coding agent. We install the CLI through uv with an...

Fireworks AI has released Fireworks Nexus, an AI management and routing platform aimed at engineering organizations. It connects the coding tools developers already use...

Microsoft AI has released MAI-Cyber-1-Flash, its first model built specifically for cyber defense. It is a 137B total, 5B active sparse MoE fine-tune of MAI-Code-1-Flash with a 256k context window. The model does not ship as a standalone endpoint — it runs inside MDASH, Microsoft's multi-model agentic scanning harness, where it handles up to 90% of tasks and pushes the system to 95.95% on CyberGym.

Moonshot AI's Kimi team and kvcache-ai open-sourced AgentENV (AENV) under MIT, as part of Kimi K3 Open Day. It runs agent sandboxes as Firecracker microVMs with millisecond snapshot, resume, and 16-way fork, behind an E2B-compatible API.

Perplexity has released pplx, an official command line client for its Search API. The tool exposes two commands — pplx search web and pplx content fetch — and returns exactly one JSON object on stdout. It ships as a checksum-verified single binary for macOS arm64 and Linux, alongside an Agent Skill for Claude Code, Codex CLI and any harness that can read a URL.

The KwaiKAT Team at Kuaishou has published the KAT-Coder-V2.5 technical report, arguing that agentic coding capability is bottlenecked by training infrastructure rather than model scale. AutoBuilder raised environment construction success from 16.5% to 57.2%, producing over 100,000 verifiable environments across 12 languages, while a sandbox audit cut RL feedback errors from roughly 16% to below 2%.

Sakana AI has released Fugu-Cyber, a security-tuned endpoint on its Fugu orchestration model. It reports 86.9% on CyberGym and 72.1% on CTI-REALM, edging past GPT-5.5-Cyber and Claude Mythos Preview. Access is gated behind manual approval, a defensive-use policy, and the Token Plan. Here is what the numbers actually mean.

Datalab rewrote Marker as a three-mode pipeline. Version 2 hits 76.0 on olmOCR-bench and sustains 2.9 pages per second on one B200 — over 5× MinerU's pipeline backend, while beating Docling on both accuracy and speed. Here's how it compares against MinerU, Docling and LiteParse, and which one fits your use case.

Today, Anthropic released Claude Opus 5. It replaces Claude Opus 4.8 as the Opus-tier flagship. Pricing is unchanged at $5 per million input tokens...

Gigatoken is an MIT-licensed Rust BPE tokenizer that encodes text at gigabytes per second. On a 144-core AMD EPYC 9565, it runs GPT-2 tokenization at 24.53 GB/s. That is 989x HuggingFace tokenizers and 681x tiktoken, against baselines already written in multithreaded Rust. The gains come from a hand-written SWAR pretokenizer and pretoken caching, not a faster BPE merge loop.

Anthropic has released the Claude Security plugin for Claude Code in beta. The plugin runs a multi-agent vulnerability scan of a repository from inside...

Cursor has made Cursor Router generally available for Teams and Enterprise plans. The system classifies each request on query, context, task complexity and domain, then routes it to the most suitable model. Cursor reports frontier-quality output at 60% savings in online A/B tests, and 30–50% savings for three early-access enterprise accounts measured against Opus 4.8 rates.

Cisco Foundation AI has released Antares, a family of small language models trained to pinpoint where known vulnerabilities live inside a codebase. Antares-1B reaches 0.209 File F1 on the new Vulnerability Localization Benchmark, above GLM-5.2 at 753B parameters and Gemini 3 Pro. The untrained Granite 4.0 checkpoints score near zero under the same protocol, so post-training supplies almost all of the capability. A full 500-task sweep runs in roughly 13 minutes on a single H100 for under a dollar, against $141 for GPT-5.5.

Poolside has released Laguna S 2.1, a 118B open-weight Mixture-of-Experts coding model with 8B active parameters per token and a 1M-token context. It matches or beats models several times its size on agentic coding benchmarks, ships under OpenMDW-1.1, and runs on a single NVIDIA DGX Spark.

Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber on July 21, 2026. The Flash tier gets cheaper and more token-efficient, with 3.6 Flash cutting output tokens 17% and dropping its output price to $7.50 per 1M. Flash-Lite runs at 350 tokens/sec, while gated Flash Cyber powers CodeMender for vulnerability finding. The flagship 3.5 Pro remains delayed.

Meta has open-sourced Astryx, the React and StyleX design system it ran internally for eight years across 13,000+ apps. It ships 150+ accessible components, seven themes, dark mode, templates, and an agent-ready CLI under MIT — with React 19+ required.

A community developer fine-tuned OpenBMB's MiniCPM5-1B on Claude Fable 5 traces into a 1B model that runs fully local — a 657MB smallest build, 128K context, and visible reasoning. We verify every spec against the Hugging Face cards, separate what a fine-tune actually inherits from real capability, and flag the licensing question the model card leaves open.

Feyn Labs has released SQRL, a family of text-to-SQL models that inspect a database with read-only probes before committing to a query. The flagship SQRL-35B-A3B reports 70.6% execution accuracy on BIRD Dev, edging Claude Opus 4.6, and distills into self-hostable 4B and 9B checkpoints.

Alibaba's Qwen team previewed Qwen3.8-Max-Preview, a 2.4 trillion-parameter multimodal MoE model it calls "second only to Fable 5." The preview is live on Token Plan, Qoder, and QoderWork at 10% of standard pricing. What is not live: any benchmark table, model card, license, per-token price, or active-parameter count. This breakdown separates what Alibaba confirmed from what it only claimed.

Google Cloud's generative-ai repository ships the Always-On Memory Agent, a reference implementation that treats memory as a running process. Built on Google ADK and Gemini 3.1 Flash-Lite, it uses no vector database and no embeddings. Instead, an orchestrator routes to Ingest, Consolidate, and Query sub-agents that read, connect, and write structured memory into SQLite 24/7.

Introduction This tutorial starts where most agent demos stop: giving the agent persistent memory, operational context, and a place to write back what happened. An...

Moonshot AI released Kimi K3 on July 16, 2026. It is a 2.8-trillion-parameter open MoE model built on Kimi Delta Attention and Attention Residuals, activating 16 of 896 experts.

OpenAI trained GPT-Red, an internal-only attacker model, using self-play reinforcement learning against a population of defender LLMs. It beat human red-teamers 84% to 13% on a replicated indirect prompt injection arena, found a novel "Fake Chain-of-Thought" attack class, and cut GPT-5.6 Sol's failures 6x on OpenAI's hardest direct injection benchmark. OpenAI concedes it still struggles with multi-turn and image-based attacks.

SpaceXAI published the Grok Build source on July 15, 2026. The Apache 2.0 Rust tree covers the agent loop, tool dispatch, the TUI, and the extension system. Grok 4.5 stays closed, and external contributions are not accepted.

Google released LiteRT.js on July 9, 2026. It is a JavaScript binding of LiteRT, Google's on-device inference library. The runtime executes .tflite models directly in the browser through WebAssembly, with XNNPACK on CPU, ML Drift over WebGPU, and experimental WebNN for NPUs. Google reports up to 3x gains over other web runtimes, and 5–60x for GPU or NPU over its own CPU path. One detail the announcement omits: tensors are manually managed and must be deleted.

See how Vibe, Claude Code, Cursor, and Codex compare on cost, open weights, self-hosting, and async agent surfaces.

OpenCoreDev has published Domain SDK 0.2.0, a TypeScript client for the custom domain lifecycle. It covers Vercel, Cloudflare for SaaS, Railway, Render, and Netlify behind one API. Status is modeled as an eight-value union, with separate verification and certificate fields.

Developer Hayden Bleasel has released Blume, an open-source, MIT-licensed documentation framework. It reads a folder of Markdown or MDX and generates a hidden Astro project, shipping static, AI-ready docs with local search, 30+ MDX components, llms.txt, and a built-in MCP server.

Anthropic's Claude Sonnet 5 narrows the gap to Opus 4.8 on agentic coding, at cheaper Sonnet token pricing.

Prime Intellect launched verifiers 0.2.0, previewing a rewritten "v1" core under the verifiers.v1 namespace. It splits an environment into a taskset (what), a harness (how), and a runtime (where), with an interception server that proxies requests and records training-ready traces. Any taskset runs under any compatible harness, with full prime-rl training support at launch.

Thinking Machines Lab published "The Future Worth Building Is Human." The essay frames human participation, model ownership, and decentralized alignment as technical challenges. It ties them to interaction models and Tinker's LoRA fine-tuning, where teams train and keep their own model weights.

MuScriptor is an open-weight, decoder-only Transformer from Kyutai and Mirelo. Trained on 170k real recordings plus 1.45M synthetic MIDIs, it transcribes full multi-instrument mixes into MIDI. The article covers its three-stage pipeline, benchmarks against YourMT3+, instrument conditioning, and setup, with an interactive explainer demo.

Robbyant, Ant Group's embodied-intelligence unit, has released LingBot-World-Infinity (LingBot-World 2.0). It is a 14B causal video generation model that behaves as an interactive world simulator. The core idea is the Mixture of Bidirectional and Autoregressive (MoBA) attention mask, paired with distribution matching distillation applied over long self-rollout trajectories. Together they target long-horizon drift, the failure mode that smears textures and warps geometry in most interactive world models. A Director-Pilot agentic harness wraps the generator, where a VLM proposes events and the Diffusion Transformer renders them. The report shows a single 60-minute uninterrupted session covering 20 scenarios. But the release is thinner than the paper: one checkpoint, a 480P reference script, no deployment code, no quantitative benchmarks, and a non-commercial CC BY-NC-SA 4.0 license.

Meta Superintelligence Labs released Muse Spark 1.1 on July 9, 2026, alongside a public preview of the Meta Model API. It is a multimodal reasoning model built for agentic tasks, with a 1,000,000-token context window the model actively compacts, zero-shot generalization to new tools and MCP servers, and multi-agent delegation across parallel subagents. Meta's own launch table shows it leading tool use while trailing Opus 4.8 and GPT-5.5 on coding.

Google AI Studio is rolling out Import from GitHub in Build mode. It transforms an existing repo into a runtime-compatible format. You can then iterate on it, deploy it, and more.

SpaceXAI released Grok 4.5, a Cursor-trained model for coding, agentic tasks, and knowledge work. It serves at 80 TPS, costs $2/$6 per million tokens, and ranks #1 on Harvey's Legal Agent Benchmark.

Netflix engineers detailed how they handle wide partitions in Apache Cassandra for the TimeSeries Abstraction. Two approaches work together: Time Slice re-partitioning tunes future partitions at the table level, while dynamic partitioning detects and splits oversized partitions per TimeSeries ID on the read path. Detection runs via byte counting and Kafka, splits are checksum-validated, and Bloom filters route reads to parallel child partitions. Average read latency dropped from seconds to low double-digit milliseconds, with 500MB+ partitions staying available.

Ant Group's Robbyant open-sourced LingBot-Vision, a self-supervised ViT family for dense spatial perception. Masked boundary modeling makes image boundaries a native training signal. The 1B backbone matches or surpasses larger models, and initializes LingBot-Depth 2.0.

NVIDIA's Nemotron-Labs-Audex-30B-A3B unifies audio understanding, speech recognition, translation, TTS, and audio generation in one MoE model. It keeps the text intelligence of its Nemotron-Cascade-2 backbone with marginal regression.

Tencent's Hy team released Hy3, a 295B Mixture-of-Experts (MoE) model that activates only 21B parameters per token. It ships under Apache 2.0 with a 256K context window, targeting reasoning, agentic, and long-context tasks. Hy3 reports 78.0 on SWE-Bench Verified and lower hallucination rates, and is free to try on OpenRouter through July 21, 2026.

OpenAI added two Realtime models to its API. GPT-Realtime-2.1-mini is a mini reasoning model for voice, priced like the earlier gpt-realtime-mini. OpenAI also cut p95 latency by at least 25% through improved caching. Here is what changed, how pricing compares, and how to connect over WebRTC.

Sakana AI has added Sakana Translate to Sakana Chat. It runs on the Namazu model series. The tool translates bidirectionally across Japanese, English, and Chinese. Three modes ship together: Translate, Proofread, and Ask. This article breaks down each mode, the XCOMET-XL benchmark, and practical use cases.

Meituan has released LongCat-2.0, a 1.6 trillion-parameter Mixture-of-Experts model that activates about 48 billion parameters per token. It pairs a native 1-million-token context, built on LongCat Sparse Attention, with training and serving run end-to-end on domestic AI ASIC superpods. Here is the architecture, the vendor-reported benchmarks, the API access path, and what remains unverified.

Most enterprise data still sits inside PDFs, scans, and slide decks. Large language models and agents cannot use that data until it becomes structured...

Anthropic released Claude Science in beta on June 30, 2026. The app runs on existing Claude models. A coordinating agent delegates to domain specialists, a reviewer agent flags and corrects citations and numbers, and every figure ships with its exact code, environment, and full message history. It manages compute across local machines, HPC over SSH, and Modal, and connects to 60+ databases plus NVIDIA BioNeMo skills.

A hands-free NVIDIA agent framework hosts each RTL problem as a versioned repository, reaching 100% completion across benchmarks.

NVIDIA's ASPIRE writes and refines robot control programs, then distills validated repairs into a reusable skill library. It gains up to 77 points on LIBERO-Pro and transfers zero-shot to unseen long-horizon tasks.

Mistral AI released Leanstral 1.5, a free Apache-2.0 code agent model for Lean 4. It saturates miniF2F and solves 587 of 672 PutnamBench problems. The 119B mixture-of-experts activates 6.5B parameters per token. We break down its architecture, benchmarks, real bug-finding case studies, and deployment code.

WebBrain is a free, MIT-licensed AI browser agent for Chrome and Firefox. It reads pages, extracts data, and automates multi-step tasks through Ask and Act modes. Run it on local models like llama.cpp or Ollama for privacy, or connect any cloud API.

Alibaba's Page Agent runs as client-side JavaScript inside the webpage. It reads the live DOM as text, then clicks and types from natural-language commands. No screenshots, no multimodal model, and no backend rewrite are required.

The Google Health API now has an open-source CLI. ghealth is a single Go binary that exposes 40 data types as agent-ready JSON. It is a community project, not an official Google release. Here's how it works, and what to check before granting OAuth access.

Anthropic is redeploying Claude Fable 5 on July 1 after US export controls were lifted. A new safety classifier blocks the technique in the Amazon report over 99% of the time, routing flagged requests to Opus 4.8. The company also proposed a four-criteria jailbreak severity framework with Amazon, Microsoft, and Google.

Google Research has released TabFM, a foundation model for tabular data. It performs zero-shot classification and regression through in-context learning. Predictions come from a single forward pass, with no per-dataset training, hyperparameter tuning, or feature engineering.

Linq launches iMessage Apps: interactive imessage_app cards that run payments, tickets, flights, and games inside the iMessage thread for agents.

OpenClaw's iOS and Android apps are companion nodes, not standalone chatbots. Each phone pairs to a self-hosted Gateway over WebSocket. This adds device hardware — camera, location, voice, and Canvas — to a local-first AI agent. Here is the architecture, the capabilities, and the trade-offs for builders.

NVIDIA's open-source BioNeMo Agent Toolkit turns biomolecular models like OpenFold3, DiffDock, and GenMol into documented, callable skills for AI agents. Each skill describes a model's purpose, inputs, artifacts, and failure modes, so an agent can select, run, and interpret it. In NVIDIA's benchmarks with Codex CLI and GPT-5.5 fast, skills raised task completion from 57.1% to 100% and doubled token efficiency.

EverMind has open-sourced EverOS, a local-first memory runtime that stores AI agent memory as plain Markdown indexed by SQLite and LanceDB. It combines hybrid BM25 + vector retrieval, multimodal ingestion, and self-evolving Skills under an Apache 2.0 license. Here's what it is, how the architecture works, where the benchmarks stand, and where it still falls short — plus a runnable code walkthrough and an interactive demo.

In this tutorial, we build a stable workflow around the Fable 5 Traces dataset from Hugging Face. We avoid fragile dependencies and manually parse the merged JSONL file to keep Colab reliable. We inspect repository files, normalize tool calls, audit structure, redact secrets, and visualize key distributions. We also export safe no-CoT chat datasets and train pure-Python Naive Bayes baselines on the traces.

Liquid AI released LFM2.5-230M, its smallest model yet. The 230M-parameter, open-weight model runs on-device at 213 tok/s on a Galaxy S25 Ultra and 42 on a Raspberry Pi 5. Built on the LFM2 architecture, it targets tool use and data extraction, beating larger models like Qwen3.5-0.8B and Gemma 3 1B on instruction following.

DeepSeek open-sourced DSpark, a speculative decoding framework that attaches a draft module to existing DeepSeek-V4 weights. It pairs a parallel draft backbone with a lightweight Markov head to cut suffix decay, then adds confidence-scheduled verification that tailors how many tokens get checked to real-time GPU load. Offline, accepted length rises 16–31% over DFlash and Eagle3; in production it speeds per-user generation 57–85% over the MTP-1 baseline, losslessly. The training repo, DeepSpec, ships under MIT.

Meta released Astryx, an open-source React design system built on StyleX. It pairs a CSS-variable theme cascade with a CLI and MCP server, so both engineers and AI agents build using the same API. The project is in Beta, MIT-licensed, and grew inside Meta over eight years.

In this tutorial, we work with NVIDIA's Open-SWE-Traces dataset to study agentic software-engineering trajectories for fine-tuning. We stream the data directly from Hugging Face, so we can process it efficiently in Google Colab without downloading everything locally. We normalize multi-turn agent conversations, parse final code patches, and build an analysis DataFrame covering trajectory length, tool usage, patch size, language distribution, and resolution outcomes. We then curate a supervised fine-tuning subset using success labels, token limits, language filters, and patch availability.

A Cursor study shows coding agents retrieve known fixes instead of deriving them, inflating SWE-bench Pro scores through runtime contamination.

Perplexity's Computer for Counsel extends Perplexity Computer to legal teams. It routes 20+ models across Midpage, MCP connectors, and Microsoft 365, with cited outputs lawyers can verify.

Apple released container 1.0, an open-source Swift tool running Linux containers as lightweight virtual machines on Apple silicon.

DeepReinforce released Ornith-1.0, an open-source coding model family built on Gemma 4 and Qwen 3.5. Instead of a fixed harness, the model learns its own scaffold during reinforcement learning. The 397B flagship reports 82.4 on SWE-Bench Verified, with all weights under the MIT license.

In this tutorial, we build a fully offline Graphify pipeline that turns a multi-module Python application into a knowledge graph. We install Graphify, generate a connected sample app, and extract the graph locally using tree-sitter, with no API key or LLM backend. We load graph.json into NetworkX and analyze file types, relationship types, centrality scores, community detection, and shortest paths. We then create static and interactive visualizations to see how modules, classes, functions, and database objects connect.

Nous Research has added /learn to the Hermes Agent Skills System. The command authors a standards-compliant SKILL.md from a local directory, a doc URL, a past conversation, or pasted notes. The live agent sources the material with its own tools, then writes the skill — no hand-writing and no separate ingestion engine. Here is how it works, where it fits, and what to review before trusting the output.

Generative AI has reshaped how software gets built. What began as line-by-line autocomplete now spans full application generation, multi-agent build pipelines, and natural-language interfaces...

UC San Diego's DFlash replaces autoregressive drafting with a lightweight block diffusion model for speculative decoding. It drafts whole token blocks in a single forward pass and conditions on target hidden features through KV injection. The paper reports up to 6.08x lossless speedup on Qwen3-8B, while NVIDIA reports up to 15x throughput on Blackwell at fixed interactivity. DFlash ships 20 checkpoints and supports SGLang, vLLM, and TensorRT-LLM.

Datalab released lift, a 9B open-weights vision model that turns PDFs and images into schema-matching JSON. It uses schema-constrained decoding for valid structure and trained abstention to return null instead of hallucinating absent fields, scoring 90.2% field accuracy on a 225-document benchmark.

Prime Intellect has released prime-rl 0.6.0, an open framework for asynchronous reinforcement learning on trillion-parameter Mixture-of-Experts models. It trained GLM-5 on SWE tasks at up to 131k sequence length, with sub-5-minute step times and 256 rollouts, on 28 H200 nodes. This breakdown covers the inference and training optimizations behind those numbers — FP8 inference, Wide Expert Parallelism, prefill/decode disaggregation, router replay, and 3-D parallelism (FSDP, EP, CP).

Fugu and Fugu Ultra route tasks across a swappable model pool, leading most coding, reasoning, and agentic benchmarks.