Change log

Junie Eco mode hands implementation to a lower-cost model

Eco mode uses the selected model to plan and a cheaper model to implement and test; JetBrains reports about half the model cost per task in internal benchmarks. It is generally available in Junie CLI 26.10.20 and later. Date is the post's publication date from the page metadata.

Sierra launches Compliance Console for outbound agents

Compliance Console, built into the Horizon platform, sets per-channel disclosures, contact windows, holiday exclusions, frequency limits, and opt-outs for agents' proactive outreach.

GitHub Copilot Fusion becomes generally available

Copilot Fusion routes and orchestrates models per task at Standard, High or Max effort, on all Copilot plans, with a limited-time 30% discount on model costs for paid plans.

GitHub Agentic Workflows become generally available

Agentic Workflows, which run coding agents inside GitHub Actions from natural-language Markdown definitions, reach general availability; the default Copilot CLI engine requires a Copilot license.

Learning Agent and admin-deployed custom MCP connectors arrive in Copilot

The release notes list a Learning Agent in Microsoft 365 Copilot on Windows and the ability for admins to create and deploy custom MCP connectors across Microsoft 365.

Glean release 565 adds agent import/export and an Azure Files connector

Release 565 adds an Azure Files connector in beta, agent import and export, 57 MCP server templates, and Gemini 4 Argon in the Model Hub, and renames chat sessions via the Platform API.

Antigravity 2.28 adds a built-in browser and any-file attachments

The Antigravity App adds a built-in browser in the side panel, a built-in PDF viewer, and attachments of any file type up to 20 MB.

Synthesia launches a Localization API for dubbing videos at scale

The Localization API translates video content into up to 20 languages through existing content systems, with per-language progress tracking.

Runway launches Runway Ads for generating and publishing ad variants

Runway Ads finds high-performing ads, creates on-brand variants, and publishes the ones the user approves.

OpenCode 2.2 adds Desktop plugins and a refreshed desktop app

OpenCode 2.2.0 lets users create, install and share experimental Desktop plugins, refreshes the desktop app, and removes the OpenCode GitHub Action.

GPT-6.1 Terra becomes available in Codex

GPT-6.1 Terra, a fast model with image understanding and computer use, becomes available in Codex and ChatGPT Work on Plus, Pro, Business, Enterprise and Edu plans.

Runway launches App Studio for building custom creative apps

App Studio builds and hosts custom creative apps from plain-language descriptions, using Runway's models and the user's brand files.

Gumloop 13.1 opens agent artifacts in Google Workspace and Microsoft Office

Release 13.1.0 lets users open agent-made documents, spreadsheets, and presentations in Google Docs, Sheets, Slides, Word, Excel, or PowerPoint, and adds live dictation.

Gumloop 13.0 revamps the agent harness and adds artifact analytics

Release 13.0.0 lets agents handle longer work and message users in chat, Slack, Teams, iMessage, and email, and adds artifact analytics and organization-wide agent evaluations.

Synthesia turns PowerPoint decks into Syren videos

PowerPoint files imported from the Home page can be converted into designed Syren videos instead of slide-by-slide scenes.

Synthesia adds management and security controls for team API keys

Synthesia's product updates list "Manage and secure your team's API keys" as a dated update.

Devin Desktop 3.11 rebuilds the Agent Window on the Devin web app interface

Devin Desktop (formerly Windsurf) 3.11.212 brings the Devin web app's sessions sidebar, Automations, Review and Wiki into the Agent Window, with side-by-side sessions and file, change and terminal views.

Glean's ClickUp connector becomes generally available

The ClickUp live-mode connector left beta and is on by default in the Admin console; existing configurations keep working.

Claude Duet opens as a research preview for Pro and Max

Anthropic opened a research preview of Claude Duet, an app in which Claude works as a proactive agent, for Pro and Max plans.

n8n 2.45 brings the Monday.com node V2 to full parity

Version 2.45 ships Monday.com node V2 with full V1 parity, makes Kafka v2 the default for new nodes, and adds Text Classifier confidence scores.

Kiro adds memory across sessions in the IDE and CLI

Kiro IDE 1.4.0 and Kiro CLI 2.29.0 add memory that carries preferences, conventions and corrections between sessions; it is off by default.

Cursor's Agent SDK becomes generally available

The Agent SDK lets teams define an agent as a folder of code, test it with evals, and deploy it to Cursor or their own infrastructure.

Perplexity adds a Decisions API and role-based connector access

The update brings Ultrafast GPT-6 Astra and GPT-6.1 Sol in Computer for Max, a Decisions API using open-weight models, and role-based connector access for Enterprise admins.

Amp becomes an agent in Linear

Assigning or mentioning Amp on a Linear issue starts a Linear Agent Session and spawns an orb; the session also appears in Amp's sidebar.

ElevenLabs launches Eleven Realtime voice agent models

Eleven Realtime is a family of voice agent models where ElevenLabs configures the LLM, speech to text, and text to speech; Realtime Translation and Video to Music also launched.

Notion AI upgrades agents, adds BYOK and a new credit system

Notion AI gets more capable agents, a cheaper Auto model router, bring-your-own-AI (BYOK) support, an improved Notion MCP, and a new credit system.

Clay adds SCIM provisioning for workspace members

Admins can connect an identity provider such as Okta or Azure AD to automatically provision and deprovision workspace members and map IdP groups to Clay roles; the roundup also added metro-area search across 151 countries.

Cline begins rolling out Cline Cloud sessions

Cline 4.1.25 starts a gradual rollout of Cloud sessions: tasks run in a hosted sandbox on a chosen GitHub repo and branch and keep running after the editor closes.

Phone calls route to teammates by skill

Inbound Intercom phone calls now ring the teammate with the matching skill first, reusing the skills already used to route chats and emails.

Personal Agent Middleware for requests from customers' AI agents

Parloa introduced Personal Agent Middleware, which applies a brand's rules and permissions when personal AI agents contact support on a customer's behalf.

Amplemarket workflows can trigger from Salesforce opportunities and HubSpot deals

Amplemarket's September 2026 updates let workflows start when a Salesforce opportunity or HubSpot deal is created or updated, and expanded the Amplemarket MCP to manage the full sequence lifecycle and pull call transcripts.

Lovable launches Growth tools for finding customers

Growth tools turn a goal such as a signup target into a plan built from the app, its audience and its traffic, from the same chat used to build.

Lindy gets its own phone number for calls

Lindy gets its own phone number for outbound calls and receptionist duties.

Cursor opens the Origin App Marketplace to public submissions

Origin integrates with over 20 apps for deployments, CI, code review, security, monitoring and project management, and the marketplace opens to public submissions.

GEO Hub adds prompt-to-page mapping; page analysis no longer uses credits

Jasper GEO Hub now runs each tracked prompt five times per AI provider, discovers pages automatically, and maps prompts to pages. Page analysis no longer consumes credits, so charges depend only on prompts and providers selected.

Notion 3.8 connects agents to workspace context

Notion 3.8 connects agents to shared Notion context through MCP, skills, version history, and column permissions.

Google Antigravity is built into the Gemini macOS app

Users can start a coding agent from chat in the Gemini macOS app and get a notification when its work is ready for review; English only, for eligible accounts.

n8n 2.44 adds Google Vertex AI credentials and renames Foundry nodes

Version 2.44 adds Google Vertex AI credentials support and renames the Foundry nodes and credentials to Microsoft Foundry.

Gemini 4 Argon rolls out in Antigravity for Google AI Pro and Ultra

Google brings Gemini 4 Argon to the Antigravity desktop app and CLI for Google AI Pro and Ultra subscribers and eligible Gemini Enterprise subscriptions.

Gemini 4 Argon arrives in Devin Desktop at 75% off through October 27

Cognition adds Gemini 4 Argon to Devin Desktop and the Devin CLI, combining Google's 50% introductory pricing with an extra 50% discount in Devin through October 27, 2026.

Cresta launches Deputy Agent, a long-running AI agent for customer operations

Deputy Agent automates work that continues after a customer interaction, executing workflows across systems and teams until the job is done.

Watch and Learn teaches Lindy a task from a recording

Lindy can learn a task from a screen recording or written steps.

Runway adds project credit caps for Enterprise workspaces

Enterprise workspace admins can set a credit budget on a project, pausing generation once it is reached; enabled through the account team.

AI Coverage puts AI Agent intents and handovers on one page

Gorgias released AI Coverage (formerly Intents), showing every intent with its tickets and handovers and letting teams choose what AI Agent handles or hands to the team.

GPT-6.1 Sol Ultrafast arrives in Codex for Pro $500 and eligible Enterprise and Edu

An Ultrafast mode speeds up GPT-6.1 Sol token generation in Codex, limited to the Pro $500 plan and eligible Enterprise and Edu plans (off by default for Enterprise).

TRAE 3.6 launches the Agent window for multi-agent work

TRAE v3.6.0 adds an agent-centric workspace for orchestrating multiple agents across projects, with cloud tasks and scheduled tasks.

Outreach October release adds Org Context for AI agents and Omni in Slack

Admins can curate company, competitor, and product context that every Outreach AI agent references; Omni became available in Slack channels and the mobile app gained Omni and SMS task improvements. Shipping October 8–27, 2026.

Context Intelligence carries customer context across conversations

Parloa launched Context Intelligence, a customer-understanding layer that lets its AI agents remember facts across conversations and channels.

Claude Code adds Claude Haiku 5.5 as its default Haiku model

Claude Code 2.1.293 adds Claude Haiku 5.5, which becomes the default Haiku model on the Anthropic API, with a 1M-token context window.

Activepieces 0.92.2 moves template expressions to the legacy sandbox

Version 0.92.2 evaluates {{ }} template expressions in the legacy sandbox instead of a Deno process per token, to relieve memory pressure on flows that map many expressions.

n8n 2.43 releases Browser Use to all users

Version 2.43 makes Browser Use available to all users and adds an Azure OpenAI deployment dropdown and a Databricks Job resource.

ChatGPT plugin, post-call emails, and a new Synthflow app

Synthflow released an official ChatGPT plugin for building voice agents, post-call email actions, and an early-access new app at app.synthflow.ai.

Perplexity adds Computer Automations and inline visualizations

The update adds Computer Automations for recurring work, inline charts and 3D visualizations, and GPT-6.1 Sol for search and Light-effort tasks for eligible paid subscribers.

Cline for Desktop cloud sessions get the same models as Cline on the web

Cloud sessions can use Cline Usage-Billing, ClinePass, and ClineFree models, and a new cloud session waits until a model is ready before sending.

WRITER MCP is live

Users can run WRITER Playbooks, pull deliverables, and bring in WRITER Meet transcripts and notes from any MCP-compatible client, using the same seat and permissions with OAuth sign-in.

lemlist adds custom enrichment waterfalls

Users can now set the order in which enrichment providers run for email and phone lookups, instead of lemlist shuffling providers randomly for each lead.

Lovable apps can add live voice conversations with GPT Live

Apps built with Lovable can add live voice conversations to their AI features with GPT Live; Lovable sets up the conversation you describe.

Codex CLI 0.160 adds opt-in Guardian review context and task history browsing

Browse older tasks in the agent command center, start sessions outside a project with workspace defaults, and opt in to Guardian review capabilities that include context from agent handoffs.

Decagon launches Voice 3, Personal Agent Gateway, Agent Modules, and Duet Apprentice

At Decagon Dialogues, the company introduced Voice 3 (with its first in-house voice model, Chord), a channel for customers' personal AI agents, industry Agent Modules beyond support, and Duet Apprentice.

Lovable Enterprise workspaces can keep AI processing in the EU

Workspace admins and owners on Enterprise plans can enable EU inference, which runs the workspace's AI model requests on providers in the European Union.

OpenCode 1.18.34 signs macOS CLI binaries with a Developer ID

Sends namespaced session identity headers with model requests and signs macOS CLI release binaries so they run reliably on newer macOS versions.

Antigravity 2.19 lets you message subagents directly

Send messages straight to a subagent from the message box, export rendered Markdown as a PDF, and revert only the conversation when you undo.

Apollo launches Builder Studio, Intelligence Layer, and Messaging OS

At ApolloNEXT, Apollo announced Builder Studio (a natural-language builder for GTM workflows), the Apollo Intelligence Layer (a GTM context engine), and Messaging OS, a signal-based execution system adding email campaigns and ad audiences.

Mission Callisto adds Gong Enrich and Deep Mode in Gong Assistant

Gong announced Mission Callisto, anchored by Gong Enrich (third-party account and contact data in the Revenue Graph), plus Deep Mode in Gong Assistant and enhancements to Agent Builder and Gong Dashboards.

Zed 1.22 adds model choice for subagents and BYOK for GPT-6.1 Sol

This week's release adds BYOK support for GPT-6.1 Sol, on-the-fly model selection for new subagents, and automatic subagent compaction.

Kiro IDE 1.2 runs multi-agent Workflows in the background

IDE 1.2 introduces Workflows, strengthens safeguards for untrusted workspaces, and adds enterprise controls for sign-in methods.

Skills roll out in Gemini chat and will replace Gems

Skills rolled out globally in Gemini chat, initially for users 18+, and existing Gems will migrate to Skills automatically.

GPT-6.1 Sol becomes available in GitHub Copilot

OpenAI's GPT-6.1 Sol is added to the models available in Copilot.

Codex CLI 0.159 lets new input steer Codex mid-response

Opt-in instant_interrupt lets new input steer Codex during model responses or long-running code-mode calls; new sessions get a compact welcome screen.

Antigravity 2.18 adds a plugin marketplace

A new Customizations tab and marketplace for discovering, installing, and managing plugins.

Claude Sonnet 5.5 added as the default Sonnet model (v2.1.284)

Release v2.1.284 adds Claude Sonnet 5.5, now the default Sonnet model on the Anthropic API.

Claude Sonnet 5.5 becomes available in GitHub Copilot

Anthropic's Claude Sonnet 5.5 is added to the models available in Copilot.

Kiro CLI 2.25 adds Power management and V3 output styles

Install and uninstall Powers from chat or the command line, choose V3 output styles, and automate session teardown with a new Hook.

ElevenLabs makes Eleven v4 and Eleven v4 Turbo available

Eleven v4 and Eleven v4 Turbo became available, alongside speech-to-text transcript editing and new agent LLM options.

Build apps for Meta devices, plus newer OpenAI and Anthropic models

The weekly update adds building for Meta devices, app creation through Muse, inline data charts, and new OpenAI and Anthropic models.

Activepieces 0.92.0 adds an AI Router and a Deno code sandbox

Version 0.92.0 adds an AI Router where one question picks the route, a Deno code sandbox, and AWS Bedrock connections with temporary SSO or STS credentials.

Amp runners can be shared with a whole workspace

Start a runner with --share and everyone in the workspace can start threads on that machine from ampcode.com.

Copilot Cowork lets users delegate complex work across platforms

The release notes list delegating complex work with Copilot Cowork on Android, Windows, iOS, Mac, and Web, along with CarPlay support for the Microsoft Copilot app on iOS.

Zed 1.21 adds BYOK for Claude Opus 5.5 and GPT-6 models

Adds BYOK support for Claude Opus 5.5 and GPT-6 Astra, Sol, and Luna, a language server command picker, and a setting that keeps the system awake during long agent turns.

Gemini CLI v0.61 hardens against indirect prompt injection

Core security hardening prevents indirect prompt injection via build file modifications and untrusted flags, and hardens filesystem boundaries.

Rollouts and Security Review bots launch on Teams and Enterprise

Rollouts monitors each pull request as it deploys and reports per-environment health; Security Review posts exploitable-bug findings on every PR.

OpenHands 1.23 ships a universal macOS desktop build

The desktop app ships as a universal macOS DMG with per-architecture bundled runtimes.

Junie CLI adds Claude Opus 5.5, GPT-6 Sol and Luna, and Grok 4.7

Junie CLI release 26.9.22 (3419.7) adds support for Grok 4.7, Claude Opus 5.5, and GPT-6 Sol and GPT-6 Luna.

v0 widens model choice, with Enterprise owner controls

An updated model picker lets users browse and add AI Gateway models; Enterprise team owners control whether custom models are allowed.

Claude Opus 5.5 added as the default Opus model (v2.1.280)

Release v2.1.280 adds Claude Opus 5.5, now the default Opus model.

AGENTS.md support: read when a project has no CLAUDE.md (v2.1.277)

Release v2.1.277: in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead.

Custom API connectors for Agent enter beta on Pro and Enterprise

Teams can connect Agent to APIs outside Replit's integration library by adding an endpoint and authentication.

Cowork becomes part of every Claude conversation

Cowork no longer requires picking a mode first, rolling out gradually to Pro and Max on web, desktop, and mobile; design, slide, and document creation is available on every plan.

Lindy Brain adds a self-updating company memory

Memory introduces Lindy Brain, a self-updating company wiki built from meetings, email, and Slack.

Warp adds native shell completions and fzf and atuin hand-off

Trigger fzf (ctrl-r or ctrl-t) and atuin from Warp, and get native shell completions alongside Warp's built-in completion specs.

Notion 3.7 adds shared agent skills and an Agent SDK beta

Teams can create, share, and run reusable AI skills; the release also puts the Agent SDK in public beta and lets Custom Agents be embedded in docs.

Activepieces 0.91.0 lets a flow step run a saved agent

Version 0.91.0 lets a flow step run a saved agent, records agent actions in the audit log, and lets an agent be configured beside the flow.

Bolt Forge, an agent on open-source models, starts a research preview

Bolt Forge joins Standard and Max as a third agent running open-source models with a larger usage allowance; it is included with individual Pro plans, and a $9/month Forge-only Bolt Lite plan opens a waitlist.

Augment's IntelliJ plugin asks before reading sensitive files

IntelliJ plugin v0.491.0: the agent asks for approval before reading .env files, private keys, certificates, and credential-named files, and MCP over HTTP now uses the configured proxy.

Cline releases Cline Desktop, an open-source app for open-weight models

Cline releases an early version of Cline Desktop, an open-source desktop app focused on open-weight models, available for Mac with a Windows beta.

ElevenLabs makes Scribe v2 Medical generally available

Scribe v2 Medical became generally available for batch transcription of medical and clinical audio.

claude plugin eval runs a plugin's eval suite with scored results (v2.1.269)

Release v2.1.269 adds `claude plugin eval` for reproducible, scored plugin evaluations.

Fusion harness comes to Devin Desktop and CLI

Cognition makes its Fusion model harness available in Devin Desktop and the Devin CLI.

v0 team chats become visible to the team by default

New chats created in a team workspace are visible to the team by default; owners can change the default visibility.

Projects launches in beta for long-running, multi-agent work

A coordinator agent plans larger bodies of work and delegates to cloud and local agents, keeping shared context across months. Rolling out to all users.

Remote agent hosts: connect to agents over SSH (v3.10.23)

SSH hosts can be saved and selected from the location picker, and sessions recover their transcript after reconnecting.

Official Make plugin for ChatGPT released

Make's release notes announce an official Make plugin for ChatGPT.

Warp adds first-class support for Grok Build sessions

Warp detects Grok Build sessions, adds branded chrome in the footer, and enables rich input mode.

Cascade is removed; Devin Local becomes the only agent (v3.9.19)

Existing Cascade conversations can be migrated with the Continue in Devin Local prompt.

Campaigns, Advanced Synthflow LLM, and business hours

Synthflow's September 4, 2026 changelog entry added Campaigns, the Advanced Synthflow LLM, Business Hours, and analytics across subaccounts.

Scheduled daily backups for production databases

Scheduled backups create one full restore point per day, with a configurable retention period.

New Invent can build agents, tools, workforces, and evals

The newest version of Invent can create and edit Agents, Tools, Workforces, Knowledge tables, Triggers, and Evals.

Granola multi-language mode expands to 32 languages

Granola now works in more languages in multi-language mode, selected under Settings > Languages.

Claude Fable 5.1 added as the default Fable model (v2.1.257)

Release v2.1.257 adds Claude Fable 5.1, now the default Fable model.

Codex in Ona follows the organization's configured model access

Inherited Codex access tokens no longer override the model access configured for the organization.

Intelligent model routing picks model and effort level automatically

Selecting Auto lets Replit route each part of the work to the model and effort level that balance quality and cost.

Cloud agents can start from scratch without a connected repository

Work is saved to a Cursor Origin repo; live previews forward to the browser, with optional publishing through Vercel.

Codex in Ona gives clearer error messages

When a run cannot continue, messages focus on what happened and what to do next, with a reference ID tied to the support bundle.

JetBrains launches free on-device Junie Local for Mac

Junie Local runs a tuned local model entirely on the user's Mac with no credits or quota and no code leaving the machine.

Cloud agents gain event subscriptions, custom modes, and per-subagent VMs

Agents can watch PRs, Slack threads, or schedules and wake on events; subagents can run on their own virtual machines.

Pika launches Pika Music for generating complete tracks

Pika Music creates complete tracks from text prompts, lyrics, voice references, or music references, alone or combined.

Maia, an AI co-worker, arrives in the Scenario Builder

Make introduces Maia, described as an AI co-worker inside the Scenario Builder.

Lindy joins Slack as a teammate

Lindy joins Slack as a teammate that works in threads and DMs.