What shipped · 2026-w45
What shipped in AI dev tools: week of Nov 2, 2026
Nov 2, 2026 – Nov 8, 2026
32 releases across 14 tools this week; biggest: GitHub Copilot (11 releases).
GitHub Copilot
@GitHub in Slack and Teams is generally available
On November 6, 2026, @GitHub in Slack and Microsoft Teams became generally available. From a conversation it can investigate bugs, create issues, implement changes, and open pull requests. In Teams, Canvas reviews implementation plans, code changes, and previews beside the conversation, and teammates can keep discussing and refining the work there. @GitHub identifies the repository from the message and starts the session without asking for that repository again. Starting is also easier when someone is not signed in, has no default repository, wants to switch repositories, or needs app access. A fix covers follow-up messages that could get no response. Issues and pull requests created by @GitHub include Created from a Slack/Teams conversation with @username in the description.
Source for @GitHub in Slack and Teams is generally available (opens in a new tab)Copilot app sessions and assisted approval
On November 6, 2026, GitHub's Copilot weekly note described changes in the GitHub Copilot app. Agent Merge can be set up from a pull request and stays enabled while required checks or rules are pending. From a pull request, reviewer status can be checked, more reviews can be requested, and selected comment text can be quoted or discussed with Copilot. When a conversation reaches its context limit, it can be compacted, or the error can show what is using the space. Dropping files on New in the sidebar starts a session. CSV and TSV files open as tables with sticky headers and an option to view the source. Pinned sessions whose pull requests have merged or closed can be archived, several sessions can be pinned or unpinned together, and nested chats can be separated without losing the conversation. My Work supports more than ten saved custom views. Personal instructions for assisted approval are opt-in, apply to local sessions, and tell Copilot which tool requests to approve without asking. The same note also points to the CLI settings dashboard, Copilot in JetBrains Air, Copilot in Xcode's coding assistant, and VS Code 1.145, which have their own notes.
Source for Copilot app sessions and assisted approval (opens in a new tab)Copilot code review settings in the API
On November 6, 2026, GitHub Copilot code review settings can be managed through the REST and GraphQL APIs, from enterprise policy down to individual repositories, on Copilot Pro, Pro+, Max, Business, and Enterprise. Reviews and comments can be identified through attribution fields. A review can report the effort it used when that metadata is available, and a comment can report the skills, MCP servers, and custom instructions Copilot named as sources. Enterprise administrators can enable or disable code review for all organizations or leave the choice to organization administrators. Organization administrators can manage enablement, the default review effort, and the runner configuration, and cannot override settings the enterprise manages. Repository administrators can manage review effort and whether reviews use custom instructions and MCP tools. The REST API accepts the same effort levels as the web interface, and organization and repository settings responses list the levels available. The enterprise and organization settings endpoints, including the organization runner endpoint, do not support GitHub Apps or fine-grained personal access tokens. Repository settings endpoints support both, with Administration read to read settings and Administration write to update them. Classic personal access tokens need repository administrator access. New review overviews keep findings at the top. Effort and surveys move into a compact footer. Skipped and content-excluded files share one Files not reviewed section, with filenames and linked exclusion reasons. Empty sections are left out, and long lists show an overflow count. After a thumbs-down on an overview, Add a reason can name the main problem, such as an approval assessment, a comment that is too verbose, or findings that are incorrect or nitpicky. Agent sessions show the AI credits a review consumed. The Automatically request Copilot code review repository rule supports the bypass list, so exempt actors do not get an automatic review. On later reviews of the same pull request, findings unrelated to recent changes are collapsed and then hidden. Balanced effort reviews use three reviewer models, and another model combines their findings. Billing and preview settings stay when code review is turned off and on again. A failed or timed-out review is retried once, and Copilot can switch to another eligible model when the preferred one is unavailable.
Source for Copilot code review settings in the API (opens in a new tab)Copilot impact dashboard adds chat and VS Code agents
On November 6, 2026, the Agent adoption chart in the Copilot impact dashboard added Copilot Chat and the VS Code Agents window, for nine Copilot experiences. The cohort formerly called Passive users is now Chat-first users: people active in Copilot who have not used a qualifying feature on at least two days in the 28-day period, so they do not qualify for an adoption phase. Enterprise and organization 28-day usage metrics reports add chat and vscode_agent to copilot_feature_engagement.totals_by_feature. Chat and the VS Code Agents window count as feature engagement and do not count toward adoption phases. Chat-first users remain the No Cohort value in the API, and cohort sizes do not change. A user counts as engaged with a feature after using it on at least two days in the 28-day period and can be counted under more than one feature. Chat covers Copilot Chat use captured by usage metrics, including agent mode, so users who make agent edits are also counted under chat. These two features, like the other seven, appear only in enterprise and organization 28-day reports. Reports for earlier periods include only the original seven features. The data is available to enterprise owners and billing managers, organization owners, and anyone with a custom organization or enterprise role that grants the View Copilot Metrics permission. The Copilot usage metrics policy must be enabled.
Source for Copilot impact dashboard adds chat and VS Code agents (opens in a new tab)Copilot usage API adds code review outcomes
On November 5, 2026, the enterprise and organization repository-level Copilot usage metrics reports added a copilot_code_review object on each repos-1-day row. total_comments_evaluated is the number of Copilot code review comments checked for whether the developer acted on them. total_comments_addressed counts those the developer acted on, by applying the suggestion, asking Copilot to make the fix, editing the code, or replying that the fix will come in a later pull request. Dividing the second by the first is the addressed rate. review_comments_by_severity, review_comments_by_review_tier, review_comments_by_comment_type, and review_comments_by_customization break both counts down by severity, review tier, comment type, and customization. total_merged_no_human_change_request_after_review counts merged pull requests that Copilot code review reviewed where no person requested changes after Copilot's first review. Comments are evaluated when the pull request merges, so they count on the merge day. Copilot usually does not evaluate comments on pull requests opened by bots, such as the Copilot cloud agent, while total_merged_no_human_change_request_after_review still includes those pull requests. pull_requests.total_copilot_applied_suggestions still counts only ready-to-apply suggestions someone applied. Counts are 0 and arrays are empty when a repository had no Copilot code review activity that day. Access requires the Copilot usage metrics policy and the View Copilot Metrics permission, or an enterprise owner, enterprise billing manager, or organization owner role.
Source for Copilot usage API adds code review outcomes (opens in a new tab)Copilot usage API flags telemetry and version gaps
On November 5, 2026, the Copilot usage metrics API added fields for users missing client telemetry or running an IDE or Copilot extension below the usage metrics minimums. On enterprise and organization user reports, client_telemetry_present is false when Copilot saw the user's activity that day but no usage events arrived from their editor, and usage_metrics_requirements lists each IDE and Copilot extension as met, unmet, or unknown against the minimum supported versions. On enterprise and organization reports, missing_client_telemetry_user_count and below_usage_metrics_minimum_user_count are that day's affected-user counts, including in each day_totals entry of 28-day reports. Each count covers a single day. A user can be in both counts, and a user can meet every version minimum and still be missing telemetry. Null or omitted values mean the data is unavailable. Days before this release have no values. Users without client telemetry still count as active, but their reports lack detail such as lines of code, feature breakdowns, and model breakdowns. Access requires the Copilot usage metrics policy and the View Copilot Metrics permission, or an enterprise owner, enterprise billing manager, or organization owner role.
Source for Copilot usage API flags telemetry and version gaps (opens in a new tab)GitHub Copilot CLI 1.0.106
Released November 5, 2026. Rex terminals render inline images using Kitty direct placements. /code opens the current session in VS Code. copilot config can set a Fusion tier, and /settings opens the unified dashboard for all users, with search across settings and local resources. The interactive CLI reports program status to supporting terminals, including when attention is needed. Opt-in session management tools and @mentions cover opted-in local sessions. COPILOT_SESSION_CWD is used when launching and expanding LSP and local MCP server configs. GPT-6.1 Terra is a selectable model under New models when it is available to the account. A Sandbox extensions setting runs extension processes inside the sandbox. MATS telemetry for Entra authentication through OneAuth respects the client telemetry opt-out. The release also includes a large fix sweep for the prompt, sandbox, MCP, sessions, and the terminal UI.
Source for GitHub Copilot CLI 1.0.106 (opens in a new tab)Copilot in VS Code 1.145
On November 4, 2026, Visual Studio Code 1.145 shipped. Updates 1.145.1 and 1.145.2 address further issues. In preview, workbench.browser.enableRemoteProxy lets the Agents window browser load pages through a remote agent host, so a dev server the agent starts there can be previewed in the Agents window. Canvases hosted on that remote host also work when chat.canvases.enabled is on. In preview, chat.agentHost.multipleChatProviders.enabled lets a nested session use a different harness than its parent, such as Claude reviewing a plan a Copilot chat wrote, or Codex given the same task. Each chat keeps its own model, agent, and permission levels. Local automations can use a five-field cron expression, which cloud automations cannot, and can choose a custom agent from the selected workspace. An experimental chat view mode, Conversation, Verbose, or Compact (the default), changes how reasoning and tool calls are shown, not the agent's context or the work it does. The Copilot harness remains the default for new sessions, can keep running in the background through the Agent Host, and can be opened from the editor and the Agents window. A Local-to-Copilot harness transition guide covers customization compatibility, session continuity, and organization rollout. In the Local harness, the Install Extension and Run VS Code Command tools, the Create New Workspace tool, and the /new slash command are removed.
Source for Copilot in VS Code 1.145 (opens in a new tab)Copilot in Xcode's coding assistant
On November 4, 2026, GitHub Copilot became available in Xcode's built-in coding assistant, in public preview starting with Xcode 27.3 beta. It requires a Mac running macOS Tahoe 26.6 or later. Setup is under Xcode, Settings, Intelligence. The existing GitHub Copilot for Xcode extension remains a separate option. All Copilot plans are supported.
Source for Copilot in Xcode's coding assistant (opens in a new tab)Copilot in JetBrains Air
On November 3, 2026, GitHub Copilot became available in JetBrains Air through the Agent Client Protocol, for tasks in a JetBrains IDE or in the cloud. Air's agent picker can use an existing Copilot subscription. Local tasks need Air installed in the IDE, GitHub Copilot selected, and the Copilot account connected. Cloud tasks run through Air Teams, from the IDE or the browser, and use the Copilot plan rather than JetBrains AI credits. A cloud task runs in a Docker container on a fresh virtual machine, traffic passes through a proxy, and no state carries over between tasks. Connecting Copilot to Air does not request GitHub Copilot permissions. Air may verify subscription or credit availability only when the cloud environment starts, and the task fails if that subscription or those credits are unavailable. You cannot choose whether an individual task uses a personal Copilot plan or an organization's plan. Air's governance and analytics features do not include Copilot usage. Air administrators can disable GitHub Copilot through a governance policy, and cannot set Copilot spending limits within Air. Cloud tasks require a JetBrains account in an Air Teams-enabled business organization, a Copilot subscription, and Copilot enabled by the Air administrator.
Source for Copilot in JetBrains Air (opens in a new tab)Copilot usage API separates BYOK
On November 3, 2026, the Copilot usage metrics API began separating bring-your-own-key usage from Copilot-hosted model usage in enterprise and organization reports, at the user and aggregate level, in 1-day and 28-day reports. Local BYOK is configured in the client: requests go from Copilot CLI or the GitHub Copilot app to the developer's provider key or a local model such as Ollama, Foundry Local, LM Studio, or llama-server. User reports add used_local_byok. User and aggregate reports add totals_by_local_byok, with session, request, prompt, and token counts by byok_kind (local_managed, local_user, local_auto, or remote_user) and surface (copilot_cli or copilot_app). Aggregate reports add daily, weekly, and monthly active local BYOK user counts. totals_by_client model rows add is_custom_enterprise_model, true when the request used an enterprise or organization custom model routed through GitHub. A model used both as a Copilot-hosted model and as a custom model appears as two entries. Existing totals are unchanged: local BYOK is a subset of totals_by_cli and totals_by_copilot_app, and custom model usage is already in totals_by_client. Local BYOK fields rely on client telemetry, so they undercount when telemetry is disabled, and some local providers do not report token usage. Custom model names that the existing normalization does not recognize appear as unknown. Local BYOK usage is not broken out by model name. Report days before this release do not include is_custom_enterprise_model. Access requires the Copilot usage metrics policy and the View Copilot Metrics permission, or an enterprise owner, enterprise billing manager, or organization owner role.
Source for Copilot usage API separates BYOK (opens in a new tab)
Lovable
GPT Image 2.6 Flare and Sunburst
On November 6, 2026, a Lovable app's AI features can generate and edit images with two OpenAI models. GPT Image 2.6 Flare (openai/gpt-image-2.6-flare) is the faster model, for previews and in-app image editors. GPT Image 2.6 Sunburst (openai/gpt-image-2.6-sunburst) is for detailed visuals and precise edits. Both can produce a short animated GIF, guide an edit with a drawing on the image, and produce a still image with a transparent background as a PNG or WebP file. Ask Lovable to use either model when it builds an image feature. Generating and editing images uses credits, like other AI usage. These models are not available while EU inference is enabled.
Source for GPT Image 2.6 Flare and Sunburst (opens in a new tab)Dreaming and publish audience emails
On November 4, 2026, Lovable started tidying project memory in the background on paid plans, a process called Dreaming. It checks its memory and the rules in AGENTS.md against the code and the project chat, then rewrites or removes what no longer applies. Rules you asked for stay as written. The app's code does not change. AGENTS.md changes show in the project chat and in version history as Lovable tidied up its notes. Dreaming does not use credits. The same day's changelog says that when you invite people outside your workspace to a published app, the audience list shows each person's email address under their name, and external invitees and collaborators are marked with an EXT label.
Source for Dreaming and publish audience emails (opens in a new tab)Abandonment protection and subfolder instructions
On November 3, 2026, Enterprise plans can protect a project from abandonment cleanup so Lovable never marks it abandoned or deletes it for inactivity. Enable Abandonment protection in the project's settings, or select Protect it on the abandoned project banner. Workspace admins and owners can protect any project in Security insights. Lovable follows AGENTS.md and CLAUDE.md files in a project's subfolders: when it opens or edits a file, it reads the instruction files in that file's folder and in every folder above it. Privacy and security settings are grouped in five tabs: Access, Publishing, Security, Integrations, and Data protection.
Source for Abandonment protection and subfolder instructions (opens in a new tab)Visitor error monitoring and BambooHR
On November 2, 2026, project monitoring on paid plans also checks errors in visitors' browsers, looks for the cause in the code, and shows what it finds in the editor. The project owner gets an email about important findings. Lovable collects browser errors only while Visitor analytics is enabled. Account preferences add Use my activity to personalize communication, enabled by default except in the EU, the rest of the European Economic Area, the UK, and Brazil. Apps can connect to BambooHR to look up and update employee records, read the company directory, and see who is out of office. Action center review dialogs show each project's owner, and a security review can ignore every open finding at once. The Publish dialog lists what stops publishing under Before you publish, and shows approval status, such as Awaiting review or Approved by with the approver's name, when the workspace requires approval. Project and workspace domain settings have a new layout, including Find and buy a domain and domain status such as Active or Needs attention.
Source for Visitor error monitoring and BambooHR (opens in a new tab)
Claude Code
Claude Code 2.1.319
Released November 6, 2026. When an MCP server runs a tool call as a task, Claude Code moves it to the background and tells Claude the result when it finishes. CLAUDE_CODE_DISABLE_MCP_TASK_BACKGROUND makes Claude wait inside the tool call instead, and CLAUDE_CODE_DISABLE_MCP_TASKS=1 turns MCP tasks off. permissions.denyInsteadOfPrompting denies a tool call that would wait on a permission prompt and lets the turn continue, in terminal sessions, subagents, and teammates. /claude-api latency-optimize looks for where Claude API code makes people wait and tries faster settings against an eval. CLAUDE_CODE_DISABLE_FEATURE_FLAGS stops feature-flag fetching without turning off telemetry, and every flag then uses its built-in default. With telemetry or feature flags off, including behind an LLM gateway, auto mode asks about or denies calls the server cannot check, with no fallback to the built-in classifier. CLAUDE_CODE_AUTO_MODE_SERVER=0 keeps the previous behavior on those setups. Bedrock, Vertex, and Foundry are unchanged. CLAUDE_CODE_AUTO_MODE_SERVER=1 turns the server-side check on behind an LLM gateway even when CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS is set. Developers behind a Claude apps gateway with no spend limit see spend for today, this week, and this month in /usage when the gateway runs this version or later. The gateway adds admin.pooled_conflict_rule and admin.pooled_priority_order to choose which pooled group budget pays first, an optional login_notice on the browser verification page and on Claude Code's sign-in screen, and CLAUDE_GATEWAY_LOG_FORMAT=json for operational logs. /tasks adds an Earlier in this session row that lists finished subagents so one can be reopened after it leaves the list or Claude Code restarts. In cloud sessions, /loop says a timer stops with its container and offers a loop that survives. The release fixes a Bash permission bypass where a crafted environment prefix could hide part of a command from permission checks, and a sandbox escape when creating a worktree. It also includes a large fix sweep for MCP, permissions, the shell, cloud sessions, Remote Control, self-hosted runners, VS Code, and the terminal UI.
Source for Claude Code 2.1.319 (opens in a new tab)Claude Code 2.1.318
Released November 4, 2026. Claude Code records your answer when it asks to turn on the mods it writes in a session. A mod's $.model.fork can set effort, so a side job can run at low whatever effort the session uses, and the result reports the effort sent. claude auth status includes loginExpiresAt for a stored claude.ai sign-in, claude -p JSON results include credential_expires_at, and claude -p text output warns on stderr starting 10 days before that sign-in ends. Mantle upstreams of the Claude apps gateway can set assume_role for a role in another AWS account and for per-developer AWS cost attribution. Claude Code no longer installs or updates the IDE extension when started in an IDE terminal; use /ide from an outside terminal or the editor's extensions view. Subagents stop instead of switching models when Switch models when a message is flagged is set to Ask each time. Background subagents move a shell command to the background after 3 minutes on all providers. deniedMcpServers URL entries also block Anthropic-provided servers whose upstream URL matches, and connectors in cloud sessions and on self-hosted runners. /claude-api managed-agents-onboard gives a new agent's built-in tools, and a quickstart template's MCP tools, the auto permission policy (Judge risky actions) by default. The 2.1.288 fix that kept the model's earlier thinking when resuming a conversation started on 2.1.286 or earlier is removed, so that thinking can be dropped. A cloud session on an Anthropic-hosted environment can start without picking a repository. The release also includes a large fix sweep for the shell, permissions, MCP, cloud sessions, Remote Control, self-hosted runners, VS Code, and the terminal UI.
Source for Claude Code 2.1.318 (opens in a new tab)Claude Code 2.1.317
Released November 3, 2026. /schedule warns when a new cloud scheduled task gets connectors you did not ask for, and says how to remove them. A check asks before a web or shell tool call sends your account email address to the network. Agent SDK hook callbacks can set onFailure to block, so a callback that fails blocks the action instead of letting it run. Claude Desktop's Code tab notices when your organization's remote managed settings have not loaded for the session, and again once they do. Auto mode notices when its check has no answer for a subagent's tool call. OpenTelemetry metrics, events, and spans add anthropic.plan, anthropic.seat_type, anthropic.provider, and anthropic.custom_base_url. The Claude apps gateway admin API adds GET /v1/organizations/analytics/cost for past spend over a date range in total, per developer, per day, or per IdP group. Gateways with admin: set keep a daily record of each developer's IdP groups for admin.identity_retention_days, which the cost report groups by. Set CLAUDE_CODE_DISABLE_NESTED_SESSION_ID to 1 on a self-hosted runner so claude -p run inside a session uses that session's ID and transcript, as in earlier versions. Subagents and teammates on a different model from the main conversation use that model's saved effort level or its default, instead of the session's. The Read tool no longer returns a file-unchanged stub; reading an unchanged file again returns its content. Cloud sessions can open or decline a link an MCP server asks you to visit. The release also includes a large fix sweep for permissions, MCP, cloud sessions, Remote Control, self-hosted runners, VS Code, and the terminal UI.
Source for Claude Code 2.1.317 (opens in a new tab)
Base44
Backend function automations deprecated
On November 8, 2026, Base44 stopped accepting new automations for backend functions. Existing automations keep running and can still be edited or deleted, but new automations entries in function.jsonc are ignored at deploy time. A function that should run on a schedule, when data changes, or when a connected tool sends an event is added as a step in a workflow instead.
Source for Backend function automations deprecated (opens in a new tab)Workspace API keys and Base code endpoints
On November 8, 2026, Base44 Enterprise workspaces can authenticate Apps API requests with a workspace API key. The key belongs to the workspace rather than a person, so it stays valid when team members leave. Owners and admins create keys in workspace settings with workspace-wide access or specific scopes, such as Deploy apps or Provision app users. The Base code API adds nine endpoints: retry and investigate failed sandbox updates, import a Claude Code or Codex session as an app branch, read a branch's git status, commit or discard uncommitted changes, open or update a pull request, and list a repository's pull requests.
Source for Workspace API keys and Base code endpoints (opens in a new tab)
Cursor
Voice chat with agents
On November 6, 2026, Cursor added realtime voice chat with agents in the Agents Window, the IDE, on the web, and in Cursor for iOS. When the agent is idle and the chat input is empty, the waveform button labeled Start voice chat sits where Send usually is, next to the microphone button for dictation. On iOS, saying Hey Siri, call Cursor starts a call with a new agent, and a Live Activity on the Lock Screen can mute or hang up. The voice agent listens, replies, and hands coding work to the agent in the chat. While that agent runs, the conversation can continue: ask what changed in a diff, check on a run, search earlier messages and calls with this agent, or switch between workers in a Project that has them. The coding agent shares short progress notes while it works and a brief answer when it is done, so the spoken update is usually shorter than the written reply. Voice chat runs on Grok Voice. Voice settings include Ara, Eve, Leo, Rex, and Sal. With Privacy Mode on, voice chat runs with zero data retention. Voice chat is separate from dictation, which earlier releases called Voice Mode. Dictation still turns speech into text in the chat input, and Cmd+Shift+Space still toggles it in the IDE.
Source for Voice chat with agents (opens in a new tab)Build live apps in Cursor
On November 5, 2026, Cursor added live apps. A cloud agent told explicitly to build an app scaffolds the web app, spins up a Postgres database if needed, sets up authentication behind Cursor SSO, and provisions hosting. If the app needs to connect to your systems, Cursor asks for the secrets it needs. Publishing picks a unique domain. Cursor runs deterministic and agentic security checks, including for secrets written in app files or exposed credentials, and nothing goes live until the checks pass and you explicitly publish. Team admins can see every app and secret on the team, add their own rules to the security review, approve publishes the review blocked, and unpublish any app. After the checks pass, the app can be shared with specific people, groups, or the whole team behind Cursor sign-in. Individual plans can also share apps publicly. A Cursor API key can be provisioned so a script or Grok Bot can use the app. Apps can be deployed from new or existing repos on Cursor Origin and GitHub. Support for other git providers such as GitLab is coming soon. The capability is available on all paid plans with cloud agents enabled.
Source for Build live apps in Cursor (opens in a new tab)
Vercel AI SDK
DeepSeek V4.1 Flash 1017 Ultrafast on AI Gateway
On November 7, 2026, DeepSeek V4.1 Flash 1017 Ultrafast became available on Vercel AI Gateway through Fireworks. Vercel says the model can generate upwards of 1,000 tokens per second. Ultrafast serves the October 17 update of DeepSeek V4.1 Flash on Fireworks' accelerated serverless endpoint, for interactive coding, reasoning, and agentic workflows that need to generate output and move to the next step quickly. It has a 1M-token context window and is in public preview, with no waitroom or session time limit. The model ID is deepseek/deepseek-v4.1-flash-1017-ultrafast. It can be selected in Claude Code, Codex, Cursor, and other coding agents after npx vercel ai-gateway setup.
Source for DeepSeek V4.1 Flash 1017 Ultrafast on AI Gateway (opens in a new tab)DeepSeek Search and Fetch on AI Gateway
On November 6, 2026, DeepSeek Search and DeepSeek Fetch became available through Vercel AI Gateway. Search returns relevant pages with titles, URLs, and summaries. Fetch retrieves page content as Markdown. Both tools work with any model that supports tool calling, using one AI Gateway API key and no separate DeepSeek account, and they stay available when the model provider changes. The helpers are in AI SDK 6.0.321 and later on v6, or 7.0.160 and later on v7, as gateway.tools.deepseekSearch and gateway.tools.deepseekFetch. If DeepSeek Search fails, DeepSeek Fetch can retrieve a result URL from Exa Search or Parallel Search. That fallback is model-directed, so provider order and retry limits are best effort. Search and Fetch each cost $0.75 per 1,000 billable requests. Failed requests, timeouts, and empty results have no tool charge. Model tokens are billed separately. Requests can use an AI Gateway API key or a provider key.
Source for DeepSeek Search and Fetch on AI Gateway (opens in a new tab)
Cloudflare Workers
AI Gateway shows provider key activity
On November 6, 2026, Cloudflare AI Gateway shows usage activity for stored provider keys, so credentials can be reviewed before rotation or removal. In Provider Keys, keys are grouped by provider with masked previews, aliases, and last-used and last-updated dates. The Key security section highlights keys whose recorded activity is older than 90 days. Those keys can be reviewed and deleted after their dependencies are checked. Usage tracking starts November 2026.
Source for AI Gateway shows provider key activity (opens in a new tab)
Codex CLI
Codex CLI 0.167.0
Released November 6, 2026. codex agents --no-daemon opens the agent command center, and the command center and session picker mark unread results. The terminal session picker can switch between connected backends and keeps drafts when switching. On supported terminals, local image attachments can be previewed while composing, and viewed local images open from clickable transcript paths. API key and cloud provider options are under Sign in another way. Enabling gcp_setup_wizard sets up Google Cloud Vertex AI with Application Default Credentials or a credential file. System proxy settings on macOS or Windows can be turned on from /experimental. Patch details can expand in the transcript while an approval prompt is open. Saved chats stay visible after a model-provider switch; if a chat cannot resume, its transcript can be read and an editable prompt prepared for a new chat. Ctrl+L clears the terminal display and keeps earlier messages available through Ctrl+T. Network approval prompts close when their requests end, and concurrent requests to the same endpoint prompt one at a time. Long realtime V3 conversations save completed speech to chat history as the transcript grows. Automatic context compaction accounts for retained output from failed responses and from responses with missing token usage data.
Source for Codex CLI 0.167.0 (opens in a new tab)
Devin Desktop
Devin Desktop 3.12.17
Released November 6, 2026. In the Agent Command Center, the session header menu lists that session's MCP servers and can turn one on or off or re-authenticate. If a local agent rejects the first prompt, Desktop puts the text back in the composer and shows the reason, so it can be corrected and resent without retyping. The editor uses VS Code 1.141.2 and Electron 43.8.3, which includes Chromium security fixes. In IDE mode, opening Devin settings opens an editor tab instead of switching modes. Chat diffs use the editor theme's diff colors, and shell commands and code blocks in the transcript use its syntax highlighting. A reopened Local session keeps its working directory and additional folders, and closing its tab no longer removes it from Past Conversations. The filename on a Local plan approval card opens the full plan before approval. Permission requests from third-party ACP agents no longer crash the conversation view or block reopening a session that has a pending request. Remote SSH connects to hosts defined only in a custom remote.devinSSH.configFile. On Windows, local edits save when the agent uses a different capitalization for the filename. Markdown previews work in the Agent Command Center.
Source for Devin Desktop 3.12.17 (opens in a new tab)
Gemini API
Gemini Embedding 2 Preview routes to the stable model
On November 6, 2026, Google deprecated gemini-embedding-2-preview. API requests to that model now route to gemini-embedding-2 at Gemini Embedding 2 rates. Update the model string to gemini-embedding-2. The stable model uses the same API schema, pricing, and embedding space, so no other code changes or vector re-indexing are required.
Source for Gemini Embedding 2 Preview routes to the stable model (opens in a new tab)
Gemini CLI
Gemini CLI 0.67.0
Released November 4, 2026. A models command and a --list-models flag list available models from the CLI. The IDE companion is bridged into container sandboxes over exec stdio. Cancellation handling stops continuation stream processing, halts auto-nudges, and prevents later tool execution in non-interactive mode. Character width is refined for terminals that render ambiguous-width characters, and the display width of Thai and Lao SARA AM is corrected. Atomic writes respect read-only files and hard links, relative file paths resolve against the active IDE workspace root, untrusted flag checks are scoped to external output, and permission requests include MCP server and tool metadata.
Source for Gemini CLI 0.67.0 (opens in a new tab)
OpenCode
OpenCode 2.2.6
Released November 6, 2026. pool: true shares one MCP connection across projects with the same setup. GUI plugins can add color themes, render or add actions to code blocks, and use native settings and menu components. On phones and tablets, composer actions sit in a touch dock above the keyboard. opencode run -w starts the session in a new git worktree, and --worktree=<name> names it. That flag cannot be combined with --continue or --session. opencode run --archived creates the session already archived, so scripted and CI runs stay out of the inbox, home, and session lists, and --continue skips them. opencode run waits for the background shell commands, subagents, and image jobs its prompt started, prints that it is waiting, and lets the model respond to their results before it exits. A job that never exits, such as a dev server, keeps the run open until Ctrl+C or SIGTERM. SIGTERM cancels the session and its background jobs and exits 143. Session warming can be set per provider and model, including interval: auto based on cache TTL. websearch.use is a policy action. OPENCODE_FLAGS=all or experimental.all_flags enables every feature flag. An experimental Code Mode bytecode VM is off by default and turns on with experimental.codemode_vm: true or OPENCODE_FLAGS=codemode_vm. Foundry Claude models can use Azure Messages. File reads and media from the server API are served as sandboxed downloads, and cross-site browser navigations and requests from disallowed origins to /api/* are refused. Webfetch and Code Mode fetch reject URLs that contain credentials, ignore the fragment when matching permission rules, and Code Mode fetch checks every redirect against webfetch rules. The TUI adds Scroll on send. Keep, session.submit_scroll set to keep, retains the reading position after a prompt is sent. The SDK adds client.experimental.job.list() for background shell, subagent, and image jobs. The release also includes a large fix sweep for the TUI, desktop, MCP, and plugins.
Source for OpenCode 2.2.6 (opens in a new tab)
Replit
Get support from Agent
On November 6, 2026, Replit Agent can troubleshoot a problem before a support ticket, in the same conversation. When asked about contacting Support, it checks for active incidents, consults documentation, and investigates known issues or platform limitations. On Pro, Agent can submit a support ticket that includes debugging details from the environment.
Source for Get support from Agent (opens in a new tab)
Zed
Zed 1.27.0
Released November 4, 2026. Adds syntax highlighting for Go 1.27 generic methods, and setting on_new_window to readme opens README files by default. AI commit messages use the three most recent repository commit messages as style examples. The Agent Panel loads tool-call diffs and parses thread message Markdown only when first viewed, and uses less CPU and fewer disk reads while the agent is working, especially with a long thread history. Agent checkpoints and the Restore Checkpoint button are removed, because taking a checkpoint before every message made the agent slow to respond, or hang, in large Git repositories. To undo the agent's changes, reject its edits or use Git. Zed now requires macOS 12.0 (Monterey) or later. The release also fixes external agents' tool-call locations opening one line below the reported line, inline review switching to a subagent's edits when the subagent started, missing thinking output from OpenAI-compatible Responses endpoints that emitted reasoning-text events, LLM credentials being read when disable_ai was true, and remote projects becoming unresponsive when loading an agent thread that referred to many files outside the project.
Source for Zed 1.27.0 (opens in a new tab)
Updated