Yesterday's Top Launches: 4 Tools from June 25, 2026

Yesterday brought a fresh batch of tools to the scene, each tackling a different piece of the modern workflow puzzle. From streamlining AI infrastructure to reimagining the creative process, these new developer tools and creative applications show a clear trend towards solving specific, gnarly problems rather than offering vague all-in-one platforms. Let’s dive into what launched.

Conduit

If you’ve built or worked with AI agents that juggle a lot of tools—think APIs, databases, various services—you’ve likely hit a wall. Performance tanks, costs creep up, and everything feels slower than it should. The culprit is often blunt-force tool discovery: every time your agent needs to think, the entire catalog of every connected server’s tools gets dumped into its context. That’s a massive token overhead before any real work even begins.

Conduit approaches this like a clever traffic controller. Instead of that overwhelming dump, it sits locally as a gateway and exposes just three meta-tools to your agent: search, call, and status. When your agent needs a tool, it asks Conduit to search. Conduit then fetches only the relevant tool schemas from its connected servers and serves them up. This on-demand retrieval is its killer feature. Early measurements show it cuts tool definition overhead by about 97% and total tokens per task by roughly 90%, all while maintaining the same success rate. For teams using cloud models, that’s direct cost savings. For those on local models, it frees up precious context window real estate for the actual task.

It’s open-source, runs on your machine, and keeps API keys in your OS keychain. It’s for developers who are past the prototype stage and are now wincing at their API bills or frustrated by agent sluggishness. The value is starkly practical.

prepros

Managing a brand shoot—whether for a product line, a campaign, or a lookbook—is notoriously fragmented. Teams ricochet between Pinterest boards, Google Docs for shot lists, PDF call sheets, spreadsheets for budgets, and frantic group chats on the day of. prepros is a workspace that tries to corral all of that into one place.

It provides structured spaces for mood boards, shot lists, call sheets, crew info, budget tracking, and even shoot-day progress tracking. The idea is to replace the collage of documents with a single source of truth. For a creative director or producer, having the mood board visually tied to the specific shots needed, which are in turn linked to the day’s schedule and the person responsible, could prevent a lot of last-minute panic. It’s less about flashy AI and more about applying solid software design to a messy, creative-adjacent process.

It feels aimed at small to mid-sized creative teams or freelancers who manage multiple client shoots. The “free options” tag suggests a freemium model, which makes sense for onboarding. Its success will hinge on whether it’s flexible enough to accommodate the unpredictable chaos of a real shoot day, or if it ends up feeling like another rigid form to fill out.

jebi

The terminal is powerful but often unforgiving. You get a cryptic error, forget a flag, or need to chain commands, and you’re off to a search engine, breaking your flow. jebi is a macOS terminal app that bakes local AI directly into the command line to smooth out those rough edges.

Its magic is in running models like Qwen or Gemma entirely on your Mac. No API keys, no data sent out. It works by offering intelligent next-command suggestions after you run something, explaining errors in plain English, and providing a chat interface via /ask. You can have it explain a git log, suggest a cleaner awk command, or debug a failing script, all without leaving the window. It also modernizes the interface with split panes, tabs, and themes.

The on-device requirement is a major plus for privacy but also a constraint; your Mac needs the horsepower. It’s ideal for developers who live in the terminal and are comfortable with AI assistants but wary of sending their shell history to a third party. It turns the terminal from a passive tool into an interactive, helpful partner.

Deckwise

Creating a presentation from scratch is a chore, and most AI tools for it are disappointingly shallow. They generate a static set of slides, and then you’re left fighting with a limited editor to make it actually look good. Deckwise starts with a different premise: what if the AI was an active co-editor throughout the entire process?

You feed it a topic, notes, or files, and it first plans an outline. Then it builds the deck on a true 2D spatial canvas—not a block-based layout—which gives the AI and you more flexibility. The standout feature is the lasso tool. You circle any element (or group of elements) on a slide and tell the AI what to do: rewrite this text, redesign this graphic, rearrange these points, polish this section. The AI acts on that specific area, considering the overall theme and layout.

This moves beyond the “prompt-and-pray” cycle into iterative refinement. It’s for anyone who needs to produce polished decks regularly and understands that the first draft is never the final version. The caveat is that it seems to rely on OpenAI’s API, so your data is processed externally. The trade-off is a potentially much more powerful and nuanced AI agent capable of complex spatial reasoning and design adjustments.

A quick recap of yesterday’s launches:

Conduit – An open-source local gateway to slash AI agent token overhead.
prepros – A unified workspace for planning and tracking brand photo shoots.
jebi – A privacy-focused, AI-integrated terminal for macOS.
Deckwise – An AI presentation agent built for interactive editing and refinement.