AI Product Updates, July 30: Cloudflare Search, Copilot, and Nine New AI Tools (2026)

July 30's verified releases were mostly about putting AI closer to the work. Cloudflare made grounded retrieval callable from common agent frameworks, GitHub expanded Copilot inside Visual Studio, and Product Hunt's launch board added tools for voice, shared memory, observability, onboarding, and personal workflows.

No frontier model release or deprecation is included here: the confirmed record for this window is weighted toward product integrations and new tools.

At a glance

Product	Releasing entity	Release date	What changed	Availability
AI Search SDK integrations	Cloudflare	July 30	Direct paths for Vercel AI SDK, LangChain, and Cloudflare Agents SDK	Available in documented SDK/package paths 1
Copilot in Visual Studio July update	GitHub	July 30	Copilot SDK Agent preview, .NET/Azure skills, selected-code review, and organization custom instructions	First three on all plans; organization custom instructions require Business or Enterprise 2
SKI	SKI	July 30 launch	Two-way voice control for several coding agents	Mac and Windows; free base product, paid AgentCall 3 4
Memmy Agent	Memmy	July 30 launch	Local-first shared memory across AI tools	Free and open source; macOS 12+/Windows 10+ 5 6
AI Search Console	AI Search Console	July 30 launch	Prompt-level AI visibility and citation tracking	Free trial; Starter/Pro/Business at $45/$195/$495 per month 7 8
Claude Code usage tracking	LangWatch	July 30 launch	Traces, token classes, cost views, and tool/MCP/file spans across coding assistants	Free individual use 9 10
Greplica	Autoloops	July 30 launch	Repo-grounded shared memory for developers and coding agents	Open source; local or managed shared mode 11 12
NINA	AgenQ	July 30 launch	Live, in-product onboarding through voice/text and screen guidance	7-day free trial 13 14
Pally	Pally	July 30 launch	AI assistant over iMessage/RCS with connected inboxes and tools	Free; Pro $25/month; Max $200/month 15 16
Expert Chase	Expert Chase	July 30 launch	E.Y.E. personal productivity assistant using connected life data	Free download on web, iOS, and Android 3 17
Yap	Frigade	July 30 launch	Private, offline macOS dictation using Apple's on-device Speech framework	Free and open source for macOS 3 18

Platform updates

Cloudflare AI Search moves into agent frameworks

Cloudflare's July 30 changelog adds direct AI Search paths for the Vercel AI SDK, LangChain, and the Cloudflare Agents SDK. The Vercel integration is a new ai-search-provider package for AI SDK v6; generated responses can return retrieved chunks as sources, and an AI Search instance can also be exposed as a tool in an agent loop. 1

The LangChain path is a CloudflareAISearchRetriever in the langchain-cloudflare package. It can run as a retriever, feed a retrieval tool, or sit inside a RAG chain, using REST credentials or a Worker binding from a Python Worker. Cloudflare's own Agents SDK could already reach AI Search through a Worker binding; the new material is the documented stateful-agent workflow that provisions an instance, indexes content, and calls search as a tool. 1

The practical change is distribution, not a new search model. Teams already using these frameworks can add grounded retrieval without hand-writing a REST client. The changelog describes integration availability, but does not announce a pricing, quota, or endpoint change. 1

GitHub adds more control inside Visual Studio

GitHub's Visual Studio update adds a Copilot SDK-based Agent in public preview, built-in .NET and Azure skills that are off by default, selected-code review, and organization-level custom instructions. GitHub says the first three are available on all plans; organization custom instructions require Copilot Business or Enterprise. 2

That combination changes where supervision happens. A developer can ask Copilot to operate through an agent surface, turn on domain skills only when needed, review a selected slice of code instead of a whole change set, and apply organization guidance centrally. The update is a workflow and governance expansion inside Visual Studio, not a new model announcement. 2

Launch-day products

SKI puts a voice layer over coding agents

SKI launched a Mac and Windows app for two-way voice interaction with Claude Code, Cursor, Codex, Gemini CLI, and OpenClaw. Its official site positions on-device speech and voice as part of the product, with a free-for-life base product and paid AgentCall participation for meetings. 3 4

The useful distinction is that SKI is an interface layer over agents users already have. It does not require choosing a new coding model; the access question is whether the supported agent can be reached through SKI's voice workflow. The launch listing marks the product as available on July 30. 3

Memmy Agent makes memory portable across tools

Memmy describes itself as a local-first, open-source memory hub for tools including Claude Code, Codex, OpenClaw, and Hermes. After explicit authorization, it scans local AI collaboration history, organizes preferences, decisions, project context, and pitfalls, then injects the relevant memory when the user switches tools. The official site says memory remains on the user's device and that scans can be revoked. Those are vendor-stated privacy properties, not an independent audit. 5 6

Memmy lists macOS 12+ and Windows 10+ access, with a free and open-source product. Product Hunt lists a free start with 2 million ChatGPT tokens. 5 6

AI Search Console measures visibility inside answer engines

AI Search Console targets SEO and GEO teams that need repeatable measurements across ChatGPT, Claude, Gemini, and Perplexity. Its launch page describes brand mentions, rankings, share of voice, competitor comparisons, cited sources, prompt-level analysis, citation gaps, and client-ready reports. 7

The product launched with a free trial and monthly Starter, Pro, and Business plans listed at $45, $195, and $495. This is launch pricing for a new product, not a documented price change from an earlier state. 7 8

LangWatch exposes the cost of coding-agent work

LangWatch's Claude Code workflow starts with npx langwatch claude and extends beyond Claude Code to Codex, OpenCode, Copilot, Cursor, and Pi. It records token classes such as cache reads and writes, separates theoretical from billed costs, and exposes tool, MCP, and file spans. Individual use is free. 9 10

That makes it more than a token counter. The detail that matters for teams is the ability to see which files, tools, and context operations are driving a coding run, then compare the model's nominal usage with the billed result. The source describes the instrumentation surface; it does not establish that every supported client exposes identical telemetry. 10

Greplica turns repository history into shared memory

Greplica is an open-source engineering memory layer. Its launch description says it extracts decisions, constraints, gotchas, failed approaches, and file-level context from coding sessions, then retrieves the pieces relevant to a task. The repository supports local SQLite mode and a managed shared mode, with install steps for coding agents. 11 12

The target is the gap between static documentation and isolated agent context. Greplica's claim is that repo-grounded memory can stay current across developers, agents, clones, and forks. “Open source” describes the code availability; the managed mode is a separate deployment path. 11 12

NINA guides users on the live interface

AgenQ launched NINA as an in-product onboarding assistant for B2B software. Users can ask by voice or text, and NINA highlights the next step directly on the live interface instead of returning a documentation paragraph. AgenQ says the SDK can be added with one line of code, and its site offers a 7-day free trial. 13 14

NINA is therefore closer to a guided UI agent than a chatbot. Its source describes workflows being taught as click-by-click paths, with the assistant moving alongside the user until the task is complete. Availability is a trial-based product offer; the source does not establish how much workflow authoring is included at each paid tier. 14

Pally brings approvals into messaging

Pally launched an AI assistant that works through iMessage and RCS, connects inboxes and tools, and asks for approval before consequential actions. The launch materials describe phone and email actions on the higher plan. Pricing is Free, Pro at $25 per month, and Max at $200 per month. 15 16

The important product boundary is the approval step. Pally is selling action through a familiar messaging surface, but the source does not imply that a message alone authorizes every downstream operation. The plan names and prices are current launch terms, not a before-and-after pricing change. 16

Expert Chase treats personal data as the assistant's context

Expert Chase launched E.Y.E., a personal productivity assistant available on the web, iOS, and Android. The company says it uses users' real-life data to answer from that data rather than from generic responses, and positions the app as a single place for tasks, calendars, habits, notes, and other personal organization. The app is offered as a free download. 3 17

The product is broad by design. Its meaningful differentiator is the data boundary: E.Y.E. is intended to work from connected personal context, while the official page does not disclose a detailed integration-by-integration permission matrix. 17

Yap keeps dictation on the Mac

Yap launched as a free, open-source macOS dictation tool for Apple Silicon and macOS 26 Tahoe. It uses Apple's on-device Speech framework, starts and stops with a shortcut, pastes into the active field, stores local transcript history, and works offline. The project says it requires no account, API key, or telemetry and that audio never leaves the Mac. 3 18

For coding-agent users, the product's value is mundane and direct: speech becomes text in a terminal, editor, browser, or message field without adding a cloud transcription dependency. “Private” and “offline” here are claims made by the project; the official page does not provide an external security audit. 18

What to take from the day

The two platform updates point in the same direction. Cloudflare is making retrieval a framework-native building block, while GitHub is adding more agent control inside an established IDE. The integration surface matters because it decides whether a capability is a separate destination or part of the tool where work already happens. 1 2

The launch-day products fill in the missing layers around that core: Memmy and Greplica carry context, LangWatch measures the run, SKI and Yap change the input method, NINA puts guidance in the product interface, and Pally adds an approval boundary before action. They are different products with different trust and deployment assumptions; “agent” is not a substitute for checking where memory lives, what tools are connected, and which actions still require confirmation.

AI Product Updates, July 30: Cloudflare Search, Copilot, and Nine New AI Tools