byteiota.com

Today · Fri 20 Nov

OpenCode Browse review for agents in your own browser: 2026 setup, security and best use

opencode browse review for agents in your own browser

AI-generated summary OpenCode released Browse on November 17, 2026, as an MCP server and Chromium extension that lets AI coding agents operate inside a user's real browser. Emergent reports that the system enables agents to work with existing tabs, cookies, and local storage while keeping sensitive steps such as logins with the user.

Also covered bybyteiota | From Bits to Bytes
Earlier:

Yesterday · Thu 19 Nov

MAI Code 2 vs. MAI Code 2 Flash in GitHub Copilot: Workload Fit, Official Token Rates, and the May 2027 Promo Window

Figure 1 Official Microsoft AI header visual introducing MAI Code 2 and MAI Code 2 Flash

AI-generated summary Microsoft and GitHub launched MAI Code 2 and MAI Code 2 Flash in GitHub Copilot with token billing and subscription tiers. Both models feature promotional pricing until May 17, 2027, and support configurable reasoning effort levels.

Also covered byAI PingByteForwardbyteiota | From Bits to Bytes
Earlier:

OpenAI Codex Security: Validate, Patch, and Verify CVEs

AI-generated summary OpenAI released Codex Security as an open-source command-line interface and TypeScript software development kit on November 12 to validate vulnerabilities, write patches, and confirm fixes. Version 0.10.0 introduces resumable scans, batch scan budgets, and clearer blocked-patch guidance for development workflows.

Earlier:

Introducing the Microsoft Copilot SDK 2.0

AI-generated summary Microsoft announced the Microsoft Copilot SDK 2.0 at Ignite, providing programmatic access to the Microsoft Copilot harness. Microsoft says the new release includes a Rust core in-process runtime that hosts many sessions within a single process.

Also covered byAzeem HassniData Phoenixbyteiota.com
Earlier:

RTK: Cut AI Coding Agent Token Bills 80% — One CLI Tool

AI-generated summary A command-line proxy tool called RTK intercepts development command output and compresses it before it reaches large language models to reduce token consumption during coding sessions. RTK supports eighteen tools including Claude Code and Cursor while claiming significant token reductions on verbose commands, though analysts note that compressed output can occasionally trigger extra reasoning loops.

OpenAI launches GPT-6.2 Astra and Sol with lower API prices and some benchmark gains

AI-generated summary OpenAI releases two new language models named GPT-6.2 Astra and GPT-6.2 Sol, featuring lower API prices and benchmark gains. OpenAI says GPT-6.2 Astra is designed for demanding professional tasks, while GPT-6.2 Sol is built for everyday use.

Also covered bymodelgrep.comAzeem Hassnibyteiota | From Bits to Bytes
Earlier:

Replit’s agent swarms built an app monitoring platform in 2 days on a $3K budget. Now, it’s offering them to customers.

AI-generated summary Replit has introduced Missions, a new product that deploys swarms of AI agents to pursue objectives over hours, days, or weeks based on user-defined parameters. Ahead of the launch, Replit says it used Missions to build an application monitoring platform named Radar in two days with a three thousand dollar budget.

Also covered byGazeWirebyteiota | From Bits to Bytes

Wednesday 18 November

Perplexity Launches Multiagent to Coordinate Frontier AI Models

AI-generated summary Perplexity introduced Multiagent for its Computer agent platform to coordinate up to 16 supporting agents across several frontier models. Claude Fable 5.2 acts as the lead, planning work and dividing tasks among models including Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon.

Also covered byMarkTechPostwebainews.combyteiota | From Bits to Bytes
Earlier:

Anthropic Scion: Git Broke at 30,000 Merges a Day

AI-generated summary Anthropic reports that its monorepo reached 30,000 merges a day in September because Claude is now writing more internal code than human engineers. Anthropic built a replacement for Git called Scion, which is an always-online client-server system that will not be open-sourced.

Earlier:

Tuesday 17 November

GPT-6 Cyber: 948 Linux Exploits, 81 Firefox Flaws

AI-generated summary OpenAI says it shipped GPT-6 Cyber on November 5 and rates the model Critical in its cybersecurity tier. OpenAI says the model found 948 working local privilege escalation exploits in the Linux kernel and 81 high-severity vulnerabilities in Firefox.

Earlier:

DeepMind AI agents prove security of compiled Signal build

DeepMind reports machine-checked proofs for a compiled Signal library build

AI-generated summary Google DeepMind researchers published a report on AlphaProof Code, where supervised AI agents generated millions of lines of Lean proof code for a compiled build of libsignal. The project links the executable to properties of PQXDH and Triple Ratchet messaging, demonstrating machine-checked proofs that verify session setup and message handling.

Also covered bybyteiota.comSchneier on SecurityThe AI Spotlight
Earlier:

GitHub Copilot retires Opus 4.8, Opus 5 and GPT-5.6 Terra

AI-generated summary GitHub retired Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra across most GitHub Copilot features on November 16. GitHub points to alternative models including Claude Opus 5.5, GPT-6.1 Sol and GPT-6 Luna.

Also covered bybyteiota.comAgentic AffairsAzeem Hassni
Earlier:

blender-ai-mcp-addon: Blender MCP Copilot Guide

Blender AI MCP Addon: Visual Chat and Multi-Agent 3D Copilot

AI-generated summary The blender-ai-mcp-addon extension runs an in-editor server enabling AI clients to read Blender scenes and author Python code. GitHub trending lists recorded about 2181 stars for the repository created November 12, 2026, indicating strong developer interest in 3D copilot tools.

Also covered bybyteiota | From Bits to Bytes
Earlier:

Monday 16 November

GitHub Copilot Rebuilds Diff View for Million-Line PRs

AI-generated summary GitHub rebuilt the diff renderer in the Copilot app to handle pull requests with over a million changed lines without performance issues. Principal Design Engineer Alberto Gimeno detailed the technical update in a November 16 post explaining how the new renderer processes code rows and comment blocks separately.

Earlier:

GitHub Enterprise Local Is GA: Run GitHub Offline

AI-generated summary Microsoft announced the general availability of GitHub Enterprise Local at Microsoft Ignite to provide regulated-industry teams with an offline-capable Copilot and repositories on controlled hardware. The product integrates Foundry Local for local AI inference and runs as a prebuilt appliance on Azure Local infrastructure.

Earlier:

NVIDIA DGX Spark Developer Kit Turns a Desk Into a Supercomputing Lab

AI-generated summary NVIDIA introduced the DGX Spark HPC Developer Kit at SC26 in Chicago for researchers, academics and developers. NVIDIA says the kit features the GB10 Grace Blackwell Superchip and scales from one node to four linked over NVIDIA ConnectX-7 networking.

Also covered byUnite.AITensor Journalbyteiota | From Bits to Bytes
Earlier: