Oracle introduced APEXlang, an abstract application-definition language, alongside the APEX AI Application Generator. Oracle says its grammar, skills and documentation will be published in the public domain so coding agents such as Claude and Codex can generate the language, while the APEX engine interprets the resulting definitions and receives monthly updates. The company is also working with OpenAI on training models to generate APEXlang more efficiently.
Meta confirmed the acquisition of Conductor, an AI coding startup whose platform deploys multiple coding agents in isolated, parallel workspaces. The agents come from external labs including Anthropic and OpenAI, and Conductor’s team is set to join Meta Superintelligence Labs. Conductor Cloud is scheduled to become read-only on November 13 and shut down on November 20, when the company says it will begin deleting cloud user data.
OpenAI launched Images 2.6 across ChatGPT, Codex, and the API, adding GIF animation, SVG export, image markup, and a complexity control for developers. The company says generation is 50% faster than the previous version, with GPT-Image-2.6 Flare producing images in as little as six seconds at low quality and Sunburst emphasizing detail and editing consistency. Hands-on testing by Zhidx found GIF generation took about 10–15 seconds but sometimes produced stuttering, artifacts, and physically implausible motion.
Kino founder Luke launched Kino on Hacker News as a code-first AI media manager and video editor for human teams and coding agents. The vendor says Kino stores footage and assets, supports timeline editing, review links and real-time collaboration, and lets Codex or Claude Code work on shared timelines through MCP access. Kino currently runs as an Electron application, with a pure web-browser version not yet released.
A Juejin tutorial asks Codex to implement a Python script that reads first-level lowercase .txt files from input/ and writes a sorted CSV containing each file’s relative path and byte size. The author reports running the script successfully and validating deterministic output, unchanged inputs, an empty directory, and a missing input directory. The first version does not recurse, delete files, or analyze text, and several filesystem edge cases remain untested.
掘金 · 人工智能 argues that agent engineering differs from conventional backend development because a model may choose its next action from newly observed material, adding a semantically uncertain decision stage. Using an Open ETF Agent v0.19.0 teaching case, the article compares tool contracts, permissions, state, recovery, testing, and delivery, then draws on a private research application to examine asynchronous delivery, human replies, memory consumption, and daily replay. It explicitly says these examples do not demonstrate real investment-research or automated-trading capability.
In a developer-authored commentary, the Mole team compares web, cross-platform and native approaches to client development, framing the choice as a trade-off among development efficiency, user experience and performance. It says AI can reduce the cost of maintaining separate native implementations, but does not remove the need for scenario-specific selection, debugging and experience validation. Mole's Mac app uses Swift and SwiftUI, while its Windows app uses C#, WinUI 3 and .NET 10. The team keeps code and AI context on a Mac and uses a Parallels Desktop Windows 11 virtual machine for compilation and testing. To keep the two implementations aligned, it converts macOS design tokens into Windows XAML resources, converts 11 localization files and maintains a SYNC.md ledger recording implemented, intentionally different or inapplicable features. These productivity and performance observations come from one project account and were not independently benchmarked.
OpenViking launched a native 3D content-understanding model that imports and parses STEP files, generating orthogonal six-view and eight-axis side-view images, summaries, model details, and indexed PMI information for downstream agent reasoning. 掘金 · 人工智能 reports that OpenViking can support agent workflows including component retrieval, BOM preparation, and supply-change analysis through integrations such as Codex and Claude. OpenViking claims 100% information recovery, but no independent testing or reproduction was provided.
The LoLBench paper introduces a benchmark for evaluating coding agents that must turn community feature proposals into cross-module implementations in large, evolving repositories. Its 100 tasks cover 29 systems, eight languages and five domains, with repositories averaging about 2.4 million lines of code. The paper reports that the best of 28 model-and-scaffold combinations resolved 14% of tasks, while all combinations had a median resolution rate of 2%.
Google has released Data Agent Kit, a free and open-source MCP toolkit that connects coding agents to more than 15 Google Cloud data services, according to InfoQ 中文 AI&大模型 citing a Google blog post. It includes guidance skills, an MCP server for live metadata and tools, and support for BigQuery Graph, Bigtable, and managed Apache Spark; it runs in Antigravity, Claude Code, Codex, Cursor, and VS Code-compatible editors. The release also separates advisory destructive-operation safeguards from IAM-enforced read-only controls, while limiting read-only enforcement to Google's MCP server and omitting commit-status and idempotency information from tool results.
OpenAI announced plans to acquire Pierre Computer Company, adding code storage, Git-based version control, navigation, and review capabilities to Codex. Pierre's Trees and Diffs libraries were already used in Codex, while Code Storage is expected to be integrated after the acquisition closes; Pierre's open-source projects are expected to remain open source. Financial terms were not disclosed, and the deal remains subject to regulatory approval.
A Juejin author prompted locally run Claude Code, Codex, Kimi, DeepSeek, Grok, and Antigravity agents for ten suggestions about the next ten years. The post reproduces outputs labeled Opus 5.5, GPT-6-Astra, and Kimi 3.1 Pro, with recurring themes including health, personal judgment, long-term focus, and systems for knowledge and automation. The author says the displayed agents broadly converged, while noting that several full outputs were omitted for space.
In a firsthand Juejin article, author 苍何 reports that WorkBuddy recently gained a browser-control feature that can open pages, click and read content inside a browser where the user is already logged in. The work happens in a separate "WorkBuddy" tab group with its own pointer, so the user can keep using other windows, and tasks can also be dispatched from the mobile client to a chosen computer. The author shows it reading official docs, filling and confirming an enterprise OA expense form, reviewing a website as a novice user, and summarizing Bilibili comments, while noting that captchas, sliders and slow or changed pages may still require human takeover.
Revolut has made self-hosted Z.ai GLM-5.3 its default AI coding model for about 2,000 engineers, running on 56 Nvidia B200 GPUs, according to TechRadar, 36氪 and The Next Web. The company reports that inference costs fell about 90%, coding time fell 60% and lead time improved 70%, while commercial models remain in use for visual and specialized tasks. Revolut says it initially imposed token limits after some engineers spent $1,000 to $10,000 per month without clear returns.
Agent Zero is an open-source AI agent with its own Linux workspace, according to the project documentation. It can use a terminal, browser and desktop applications; users can monitor it, take over with their own mouse and keyboard, and connect it to local files when needed. The documentation recommends A0 Launcher for desktop computers and A0 Install for servers or shell-only environments. These capabilities were not independently tested in the supplied report.
The blader Humanizer repository describes an installable agent skill that rewrites AI-generated text to sound more human without changing what it says. It is built on Wikipedia’s “Signs of AI writing” guide, checks for 26 writing patterns and supports voice matching. The project says it works with Claude Code, Codex and other agents that support skills, including Gemini CLI, GitHub Copilot and Windsurf; these compatibility claims have not been independently verified. The project also reports that judges preferred its rewrites in all 16 blind-test comparisons, a result that has not been independently validated. It says the tool is intended for human readers rather than AI-detector evasion, and that detectors still flag most output.
Spotify's Portal AI Plugins repository supports Claude Code, Codex, Cursor, and pi. Its README says the workflows configure Portal CLI authentication, search software catalogs and technical documentation, generate operational service briefings, run read-only diagnostics, and invoke Portal actions with dry-run and confirmation safeguards. It also documents Fleetshift workflows for creating, running, checking, and troubleshooting shifts and opening pull requests after confirmation. A separate shunt plugin, described as Claude Code-only for now, routes I/O-heavy work to AiKA modes using cheaper worker models. The available evidence is the repository README, and its capabilities have not been independently verified.
掘金 · 人工智能 presents Magpie v0.1.2906 as an open-source local gateway and configuration manager for AI coding agents. It converts among OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and Gemini APIs, allowing Claude Code to use DeepSeek and Codex to use Kimi while managing provider keys, model settings and reusable profiles. The report also describes routing groups, intent-based model selection and local key redaction, but its functionality claims come from one unofficial tutorial citing project documentation.
OpenAI launched GPT-6 Cyber, a cybersecurity model for discovering, validating and developing exploits for vulnerabilities within authorized research scopes. OpenAI said an associated prototype analyzed 2,308 candidate Linux kernel findings, reproduced 1,724 vulnerabilities and produced 948 working local privilege-escalation exploits; separate reports attributed an ExploitGym score of 48.5% and 81 previously unknown Firefox vulnerabilities to the model. Initial access is limited to approved Daybreak Red customers through the API and Codex, with the model’s cybersecurity risk rated Critical.
InfoQ 中文 AI&大模型 reported that OpenClaw founder Peter Steinberger estimated at Cloudflare Connect that his AI work consumed about $10 million in tokens the previous month while running roughly 100 Codex agents. The report also examined disputed estimates for engineer Lauren Tan’s Cursor usage, ranging from about $132 million annually under an $8-per-million-token assumption to $615,000 per month using an observed Opus 5.5 average, while stressing that model mix, caching, discounts, and pricing can change the results by orders of magnitude.
An author reports that repeated attempts to buy Claude Pro through a newly registered U.S.-region Apple account failed, including with a gift card, while purchases of other apps worked. The author gave Codex the problem details and requirements, and Codex operated the computer and browser to open Apple Support. The report is inconsistent about the agent’s identity: it says Codex handled the browser workflow, but later describes the customer-support conversation as being conducted by GPT. Apple Support verified the account, escalated the case for review, asked the user to wait 72 hours before retrying, and said no review-confirmation email would be sent. The customer-service representative did not disclose the failure’s cause or backend adjustments. The author ultimately reported that the subscription succeeded.
Security researcher @log4j said OpenAI deactivated his account for alleged “Cyber Abuse” after he used ChatGPT to analyze code on his own 2025 Samsung television, reporting that roughly $500 in balance was also lost. According to 36氪 and cnBeta, OpenAI restored access and apologized after 11 days, while OpenAI researcher roon said Codex had never prohibited reverse engineering and the enforcement issue was being fixed.
xAI’s Grok Bot is reported to have launched as a personal agent team on August 11, 2026, with role-specific bots for tasks such as research, writing and scheduling. The bots can work in parallel, message one another, hand off tasks, maintain separate memories, and operate in cloud-computer environments; the reports also cite 800,000 paid weekly users, although that figure and other growth claims were not independently verified. 硅星人Pro’s product review says the product’s distinction comes primarily from integrating existing model, memory, cloud-computer and multi-agent capabilities into a product aimed at everyday personal work.
A tutorial maps six coding-agent primitives—AGENTS.md/CLAUDE.md, Skill, Subagent, Hook, MCP, and permissions with sandboxes—to the five stages of AI-driven software development. It proposes storing roles, skills, and hooks in neutral source files, then generating host-specific projections for Claude Code and Codex. The article also describes differences in read-only subagents, hook approval, and skill directories, alongside preflight and log-based verification.
A Juejin article argues that AI Skills are reusable capability packages rather than complete products, making their standalone pricing difficult. It contends that purpose-built Bots offer clearer value because they deliver finished results and can support recurring, usage-based or outcome-based charges. The article frames Skills as infrastructure or components within broader Bot products.
Author Cang He summarizes the DeepSeek Harness team’s Super-agent design, including parallel Workers with independent contexts, direct delivery of follow-up messages through subscription channels, database-backed session state, and an external Watchdog for monitoring execution and exceptions. He then describes his own LobeHub workflows for delegation, recurring inspections, local Codex and Claude Code connections, code review, and acceptance evidence. The account also documents unresolved document-access and review-configuration issues and says that the initial review was not independently verified.
IT之家, cnBeta and TMTPost report that ByteDance has introduced restrictions on using personally funded Claude and Codex accounts to access L4 code repositories and Feishu documents. A source cited by 36氪 says public work information and non-work personal use remain unaffected. ByteDance technical vice president Hong Dingkun attributed the change to the serious consequences of leaking L3-and-above content, while saying other tools should not be blocked merely to promote ByteDance products.
OpenAI featured DevRev’s Computer and Operator, which combine OpenAI models with a harness for company knowledge, permissions, and controlled agent actions. DevRev says the agents can reason across live data, reverse their actions, and cite source answers while handling customer and employee requests. GPT-6.1 Sol and GPT-6 Astra are used across these workflows, while DevRev’s engineers have also standardized on Codex.
Baseten announced three infrastructure releases: Grid enters public preview as a gateway, intelligent router, and guardrail layer for coding models; Sandboxes enter public preview as stateful, VM-isolated agent environments; and Loops becomes generally available for serverless supervised fine-tuning and reinforcement learning. The vendor said Grid is compatible with tools including Claude Code, Codex, OpenCode, and Pi, while Loops supports Tinker-compatible workflows and direct deployment through Dedicated Inference. Baseten also cited OpenEvidence, Datadog, and Opendoor as customers using different combinations of its training, inference, and routing products.
In a first-person account, the author says they moved from engineering on Microsoft’s Windows AI team to research engineering at Salesforce AI Research, with more time spent supporting scientists and rapid prototyping. They now report managing about five to six Claude Code, OpenCode, Codex, or Pi agents for coding, research, writing, and creative work, while drafting a personal AI policy to guide daily use.
What problem it solves
Developers using AI coding CLIs need a way to delegate work among multiple agents and coordinate their execution. CAO addresses this by running provider CLIs in isolated terminal sessions and giving a supervisor tools for coordinating workers.
Key features
Multi-agent delegation — A supervisor delegates work to specialist agents, with parallel and sequential execution suppor…
What problem it solves
Dodaios provides a locally installed, open-source way to make a personal AI assistant the control center for work on your computer. Its lead assistant coordinates models, coding agents, and computer use, while chats, memory, documents, and settings remain files on the computer that runs Dodaios.
Key features
Voice, chat, and phone interaction — The lead assistant supports wa…
What problem it solves
Web teams and developers can need to rebuild an existing site in a modern codebase when migrating platforms, recovering lost source code, or learning how a production site is constructed. This template gives an AI coding agent a repeatable workflow for recreating a site as an editable Next.js application while comparing its appearance and behavior with the source.
Key featur…
What problem it solves
Developers using multiple AI command-line tools need a consistent way to manage providers, accounts, prompts, MCP servers, sessions, and configuration. Switching configurations manually across applications can also make it difficult to inspect usage, preserve settings, or use different providers in separate terminal sessions.
Key features
Provider management — Manage API con…
What problem it solves
Shell users need a searchable, durable record of commands and terminal activity across machines, while AI agent users need continuity between sessions and harnesses. Atuin addresses both needs by recording terminal and agent activity and making it available for search, resumption, and optional synchronization.
Key features
Searchable shell history — Atuin records terminal ac…
What problem it solves
Coding agents can generate code without consistently applying a disciplined engineering workflow. Without reusable processes, teams may have to rebuild planning, testing, review, security, and learning practices in every prompt.
Key features
Coordinated engineering workflow — The system organizes agent work into a repeatable cycle that moves from planning and testing through…
What problem it solves
The project brings market monitoring, watchlists, portfolio records, research, news, backtesting, and report management into one workbench across China A-shares, US stocks, and HK stocks. It also keeps research records and imported materials on the user's computer while allowing research to be conducted with local AI subscriptions or compatible model APIs.
Key features
Unifi…
What problem it solves
AI agents often produce plausible SQL without the business definitions, approved joins, units, and examples needed to make results trustworthy. Teams also need a reviewable way to manage that context and turn governed answers into shareable dashboards across existing data sources.
Key features
Governed semantic and context layers — Business meaning is represented through the…
What problem it solves
Personal and team knowledge is distributed across notes, conversations, trackers, and research, making it difficult to capture, organize, verify, and synthesize useful information. COG provides a markdown-based workflow in which agents can maintain a shared knowledge base, support recurring research and operational tasks, and verify selected work through independent checks.
…
What problem it solves
Individual agents have different capabilities, context, workflows, and ecosystem connections, forcing users to switch interfaces, repeat information, and coordinate subtasks manually. Users need a way to delegate a task to appropriate installed agents and consolidate their work without managing each agent separately.
Key features
Automatic agent selection — AutoSwarm selects…
What problem it solves
Configurations for the Codex desktop app, CLI, API providers, prompts, Sessions, Skills, MCP, and TOML files can become scattered across multiple locations and formats. Codex-X consolidates these recurring configuration tasks into a visual desktop interface so the current state is visible and common operations can be completed directly.
Key features
Visual prompt management …
What problem it solves
Switching AI tools should not mean rebuilding or manually copying your setup. Teams also need a controlled way to version, share, audit, and distribute skills and other AI resources across projects and machines.
Key features
Unified resource management — Keep skills, agents, rules, MCP connections, hooks, and plugins in one place, then distribute them to the AI tools you use…
What problem it solves
AI agents need product-specific instructions to use NVIDIA software reliably across physical AI, robotics, simulation, CUDA-X libraries, RAG, and platform workflows. The catalog centralizes these skills while keeping them maintained in their respective product repositories and synchronized through an automated pipeline.
Key features
Agent-discoverable skill catalog — The NVI…
What problem it solves
Teams using coding agents need a shared way to access conversations, dispatch work, and follow activity across machines and surfaces. Screenshots, pasted logs, and separate agent sessions make context difficult to preserve and coordinate.
Key features
Shared agent conversations — Open the same conversation with teammates and follow its transcript, runtime status, files, and …
What problem it solves
Publishing videos, articles, and image posts across social media requires different platform-specific workflows and account preparation. Creators and operators need a consistent way to manage sessions and publish supported content through command-line and Agent-compatible interfaces.
Key features
Multi-platform publishing — Supports video, article, and image-post publishing …
What problem it solves
ChatGPT conversations need a controlled way to work with real project files, terminals, desktop applications, and long-running tasks. Users also need visibility, intervention, recovery, and session continuity when work is delegated across workers or spans multiple chats.
Key features
Local project and terminal control — ChatGPT can read and edit files, run tests, keep termin…
What problem it solves
Creating a complete character animation set requires more than generating isolated images: each action must be keyed, registered, timed, finished, and packaged for the target engine. Pixel-art workflows also need methods that convert model-generated flat artwork into consistent sprites on a real grid.
Key features
Complete sprite-set pipeline — Generate an approved master st…
What problem it solves
AI coding tools often expect particular provider protocols, making it difficult to use the models and providers a user prefers. A local gateway can translate between supported client APIs and provider APIs while preserving capabilities such as streaming, tool calls, reasoning, and images.
Key features
Provider protocol translation — The gateway translates Responses, Chat Com…
What problem it solves
Developers and coding agents need to understand how code is connected before making changes. GitNexus provides a graph-based view of symbols, callers, dependencies and execution flows to support code discovery, impact analysis and refactoring.
Key features
Graph-based code intelligence — Index a codebase as a knowledge graph covering symbols, calls, dependencies and executio…
What problem it solves
Agent memory is often opaque: stored context cannot be easily inspected or understood. OpenViking addresses this by organizing knowledge, memory, and skills in a browsable virtual filesystem, with summaries that help agents identify relevant context before reading full files.
Key features
Unified context filesystem — Knowledge, memory, and skills share one filesystem address…