New AI Tools Weekly #6: Agents become product surfaces (2026)

The week's best signal is that agents are no longer just chat windows wrapped around models. They are turning into product surfaces: video editors, deployment platforms, customer-context canvases, local memory vaults, QA runners, and workspaces where multiple coding agents can run side by side.

The strongest launches and trending repos from June 23-29, 2026, cluster into five patterns. GitHub was again the densest source for open tooling; Product Hunt added the clearest SaaS launch signals. Public X search was noisy this week, so I did not promote X-only candidates without a stronger Product Hunt, GitHub, or official-page trail.

Quick scan

Theme	Tools to try first	What changed this week
Media agents become production systems	OpenMontage, Palmier Pro, Voicebox, Folio AI	The interesting tools are packaging editing, voice, slides, and video as repeatable workflows rather than one-off generation.
Agent deployment gets productized	Tencent EdgeOne Makers, AWS Agent Toolkit, Hyperframes	Cloud and rendering infrastructure is being presented directly to agents, not only to human developers.
Memory and context move into user-facing tools	note.md, Cognee, Propane, design.md	Context is becoming a file, graph, or workspace that agents can reuse across sessions.
Agent workspaces split into fleets and UI control	Orca, Agent Native, Page Agent, AI Website Cloner Template	The coding-agent market is shifting from single-agent prompts to orchestration, app frameworks, and browser/page control.
Evaluation and vertical agents get narrower	Agent Arena, QApilot CoWork, Hiring Agent	The better products define a specific evaluation loop: competitions, mobile QA, or resume scoring.

Media agents become production systems

Tool	Source signal	Pricing / access	Differentiation	Try-it recommendation
OpenMontage	GitHub's weekly all-language trend snapshot listed OpenMontage with 27,683 total stars and 18,703 stars in the window; the repo describes an open-source agentic video production system with 12 pipelines, 52 tools, and 500+ agent skills.GitHub Trending weekly OpenMontage repository	Open-source repo.	It treats video production as a pipeline inventory for coding agents, not just a prompt-to-video interface.	Try if your team already uses coding agents and wants a reproducible video workflow with inspectable steps. Skip if you only need quick social clips.
Palmier Pro	The weekly GitHub snapshot showed Palmier Pro at 9,395 total stars and 5,034 stars in the window; its README calls it "the video editor built for AI" and says it requires macOS 26 on Apple Silicon.GitHub Trending weekly Palmier Pro repository	Public GitHub repo; macOS app download path in README.	Its bet is local, AI-aware editing on the desktop rather than a cloud-only video generator.	Try if you are on Apple Silicon and want to see whether AI-assisted editing feels better inside a native editor.
Voicebox	GitHub's TypeScript weekly trend list showed Voicebox with 35,590 total stars and 3,883 stars in the window; the README describes a local voice I/O stack for cloning, speech generation, dictation, and talking to agents in owned voices.GitHub TypeScript Trending weekly Voicebox repository	Open-source repo.	Voice is handled as local agent I/O, not a hosted voice-demo wrapper.	Try if privacy, local dictation, or agent voice UX matters. Check responsible-use constraints before testing cloned voices.
Folio AI	Product Hunt's June 27 leaderboard ranked Folio AI first for the day; the product page says it works in PowerPoint and Google Slides, with "Free Options" listed on the launch page.Product Hunt: Best of June 27, 2026 Folio AI on Product Hunt	Freemium / free options.	It is aimed at editing existing slide workflows, not replacing decks with a separate AI presentation canvas.	Try if your slide work already lives in PowerPoint or Google Slides. The vendor's 6x speed claim is worth testing on a real deck before trusting it.

Agent deployment gets productized

Tool	Source signal	Pricing / access	Differentiation	Try-it recommendation
Tencent EdgeOne Makers	Product Hunt's June 24 leaderboard ranked Tencent EdgeOne Makers first; its product page describes an edge platform for modern web apps and AI agents, with runtime, sandboxed tools, memory, observability, model-gateway support, serverless functions, and storage. Product Hunt lists it as free.Product Hunt: Best of June 24, 2026 Tencent EdgeOne Makers on Product Hunt	Free launch listing.	It packages deployment, runtime, storage, and observability around agents. That is closer to "agent app hosting" than a generic CDN pitch.	Try if you need to ship an agent-backed web app quickly and want hosting primitives in one place. Compare lock-in and region fit before moving production workloads.
AWS Agent Toolkit	GitHub's Python weekly trend list showed AWS Agent Toolkit with 1,565 total stars and 600 stars in the window; the README says it helps AI coding agents build, deploy, and manage AWS applications and works with Claude Code, Codex, Cursor, and Kiro.GitHub Python Trending weekly AWS Agent Toolkit repository	Open-source AWS-supported repo; Apache 2.0 badge in README.	AWS is formalizing guardrails, plugins, skills, and service knowledge for coding agents instead of leaving teams to hand-roll prompts.	Try if your agent writes infrastructure changes. Start with a non-production AWS account and watch what permissions the agent receives.
Hyperframes	GitHub's TypeScript weekly trend list showed Hyperframes with 32,057 total stars and 2,650 stars in the window; the repo tagline is "Write HTML. Render video. Built for agents."GitHub TypeScript Trending weekly Hyperframes repository	Public GitHub repo.	It gives agents a web-native way to describe video compositions through HTML, which is easier to diff and review than opaque timeline state.	Try if your product needs automated video rendering that developers can inspect in code. Skip if you need a human-first nonlinear editor.

Memory and context move into user-facing tools

Tool	Source signal	Pricing / access	Differentiation	Try-it recommendation
note.md	Product Hunt's June 26 leaderboard listed note.md third; the product page describes a local-first macOS research workspace for papers, citations, markdown notes, and LLM memory. Product Hunt lists "Free Options."Product Hunt: Best of June 26, 2026 note.md on Product Hunt	Freemium / free options.	It connects citation management and local notes to agent memory. That is more useful than a plain markdown editor if your work depends on evidence trails.	Try if you write research-heavy docs on Mac and want local files to become agent context. Skip if you need a collaborative web workspace today.
Cognee	GitHub's Python weekly trend list showed Cognee with 25,148 total stars and 6,064 stars in the window; the repo describes an open-source AI memory platform using a self-hosted knowledge graph engine.GitHub Python Trending weekly Cognee repository	Open-source repo.	It focuses on persistent memory across sessions, with a graph model that can be self-hosted.	Try if your agents repeat retrieval work across sessions. Benchmark ingestion and retrieval latency on your own corpus before replacing a simpler vector store.
Propane	Product Hunt's June 24 leaderboard ranked Propane second; its product page says it creates an always-current customer context view for product teams and agents, collected from connected tools and committed to coding or design agents. Product Hunt lists "Free Options."Product Hunt: Best of June 24, 2026 Propane on Product Hunt	Freemium / free options.	It turns customer context into a shared canvas that agents can act on, which is different from a passive analytics dashboard.	Try if PMs and designers keep re-explaining customer context to coding agents. Watch permissions and data retention before connecting production customer systems.
design.md	GitHub's TypeScript weekly trend list showed design.md with 23,012 total stars and 6,728 stars in the window; the repo describes a format specification that gives coding agents a persistent, structured understanding of a visual identity.GitHub TypeScript Trending weekly design.md repository	Public GitHub repo / specification.	It is not an app; it is a context file for design systems. That makes it useful precisely because it can travel between agents and codebases.	Try if agents keep breaking brand rules or rebuilding the same UI choices. Start with one product surface and compare outputs before and after the spec.

Agent workspaces split into fleets and UI control

Tool	Source signal	Pricing / access	Differentiation	Try-it recommendation
Orca	GitHub's TypeScript weekly trend list showed Orca with 8,821 total stars and 2,769 stars in the window; the repo describes an ADE for working with a fleet of parallel agents, available on desktop and mobile.GitHub TypeScript Trending weekly Orca repository	Public GitHub repo.	It assumes developers will run several coding agents at once and need an environment to manage them.	Try if you already parallelize tasks across agents. Do not start here if your team is still learning how to review one agent's patch safely.
Agent Native	GitHub's TypeScript weekly trend list showed Agent Native with 2,962 total stars and 1,540 stars in the window; the repo describes a framework for building agent-native applications.GitHub TypeScript Trending weekly Agent Native repository	Public GitHub repo.	It treats agents as an application architecture choice, not a bolt-on chat feature.	Try if you are starting a new product surface where agent state, actions, and UI need to be designed together.
Page Agent	GitHub's TypeScript weekly trend list showed Page Agent with 20,537 total stars and 1,778 stars in the window; its README says it is a GUI agent living in a webpage that controls web interfaces with natural language.GitHub TypeScript Trending weekly Page Agent repository	Public GitHub repo; MIT badge in README.	It embeds the control layer inside the page, which can be cleaner than trying to infer UI state from screenshots alone.	Try if your web app has repetitive flows users would rather describe than click through. Keep destructive actions behind confirmations.
AI Website Cloner Template	GitHub's TypeScript weekly trend list showed the template with 23,176 total stars and 5,317 stars in the window; the repo description says it can clone any website with one command using AI coding agents.GitHub TypeScript Trending weekly AI Website Cloner Template repository	Public GitHub template.	Its value is not originality; it packages the prompts, commands, and agent folders needed for a specific web-rebuild task.	Try for internal prototyping or migration spikes. Do not use it to copy protected brands or live products.

Evaluation and vertical agents get narrower

Tool	Source signal	Pricing / access	Differentiation	Try-it recommendation
Agent Arena	Product Hunt's June 26 leaderboard ranked Agent Arena first; the product page describes an open competition network where autonomous agents compete in real-world challenges, earn rewards, and build reputation. Product Hunt lists it as free.Product Hunt: Best of June 26, 2026 Agent Arena on Product Hunt	Free launch listing.	It is closer to an agent benchmarking marketplace than another internal evaluation dashboard.	Try if you build agents and need external challenge data or public reputation signals. Treat reward dynamics carefully; benchmarks can be gamed.
QApilot CoWork	Product Hunt's June 27 leaderboard ranked QApilot's CoWork second; its product page says it converts existing test cases into mobile automation with AI planning, human-approved replanning, and real-device execution across iOS, Android, and Flutter. Product Hunt lists "Payment Required."Product Hunt: Best of June 27, 2026 QApilot CoWork on Product Hunt	Paid.	It focuses on the high-friction mobile QA loop where brittle scripts, device variance, and human approval all matter.	Try if your mobile QA backlog is large but your test cases are already well written. Skip if your team has not cleaned up manual test definitions.
Hiring Agent	GitHub's Python weekly trend list showed Hiring Agent with 3,349 total stars and 1,973 stars in the window; the README describes a resume-to-score pipeline that extracts PDF data, enriches it with GitHub signals, and produces fair, explainable evaluations.GitHub Python Trending weekly Hiring Agent repository	Public GitHub repo; MIT badge in README.	The interesting part is explainability and enrichment, not the generic claim that AI can screen resumes.	Try only for internal process experiments with human review. Do not use it as an automated rejection system without legal and bias review.

What to do with this week's stack

If you only have one hour, test the tools in this order:

For builders shipping agents: Tencent EdgeOne Makers, AWS Agent Toolkit, and Agent Native cover deployment, cloud guardrails, and app architecture.
For teams drowning in context: note.md, Cognee, Propane, and design.md attack the same problem from different sides: research memory, graph memory, customer context, and brand context.
For media workflows: OpenMontage, Palmier Pro, Voicebox, Hyperframes, and Folio AI are worth a hands-on trial because their value depends heavily on workflow fit.
For quality and evaluation: Agent Arena, QApilot CoWork, and Hiring Agent should be tested with strict guardrails. They touch evaluation, QA, or hiring decisions where false confidence is expensive.

The pattern is clear enough to act on: the useful AI tools this week are less about "a smarter chatbot" and more about the surrounding machinery that lets agents remember, deploy, render, test, and operate inside real workflows.