Yesterday's Top Launches: 5 Tools from September 25, 2026

Five products landed on September 25, and the batch leans hard toward software that carries more of the work itself. Two of them hand whole jobs to autonomous agents, one gives a team a shared memory across every AI tool it touches, and two chip away at the daily friction of typing and talking. Only a couple fit the traditional definition of new developer tools, but all five assume you would rather describe an outcome than click through the steps to reach it.

Solid

Solid is the most ambitious thing in yesterday’s pile. It gives AI agents their own computers, their own accounts, and a budget, then turns them loose on a job described in plain language. The agents work out the steps, connect whatever tools the job needs, build anything missing, and check the result before reporting back. No prebuilt connector is required, which matters when the tool you need has never been added to an integration directory.

The self-healing and self-scaling mechanics are where it gets interesting. Agents repair broken setups, learn from your team’s corrections so the next job starts smarter, and can spawn more agents (or call in Codex and Claude Code) when a task outgrows one worker. A meta-agent layer lets you ask what is running, why, and what it cost. Budgets and approval rules cap the spending, so agents can research and draft freely while purchases and deployments wait for a human. Pricing runs $40, $160, or $640 a month, with each subscription becoming a single balance for AI usage, machines, and purchases. The roster of users it lists, from Revolut to the Government Digital Service, is impressive, though an agent that signs up for services and spends your money is exactly the kind of promise that needs testing before you trust it with a production fix.

ToneBird

ToneBird is narrower and, for that reason, easier to trust. It is a desktop reply assistant for Mac and Windows that drafts answers inside Gmail, Slack, WhatsApp, and a dozen other apps. What separates it from a generic autocomplete is memory tied to relationships. Person cards hold the running context for each contact, so it knows that Sam’s scope changed last quarter or that David wants a checkpoint at 3 PM, and it pulls agreed prices straight out of connected proposal files.

It drafts three options, adapts tone per person, and keeps your voice intact through a profile with sliders for directness, formality, and warmth. It never sends for you. There is a follow-up list that surfaces promised quotes and unanswered questions, plus calendar scheduling once a time is agreed. The free tier gives 15 uses a day and 30 contacts; Standard is $9.90 a month. The scope is the reply rather than the document, which is a sensible limit.

Lightmeter

Lightmeter takes a stance. It is an iOS app that doubles as a real light meter when you carry film and a film camera when you do not, and it applies zero AI processing and no HDR. Instead it captures true RAW and models grain, halation, and color on top, with twelve looks spanning color emulsions and monochrome stocks like Mono X400. The claim is that the film character is baked into capture and development rather than bolted on as a filter.

The privacy posture is refreshing: no signup, no tracking, no ads, nothing leaving your phone. The metering mode is the detail that will matter most to analog shooters, since it means one app covers both habits. No pricing is listed, and the anti-AI position is a bet that a specific audience wants grain over sharpness.

GBrain

GBrain gives a team one shared AI memory that stays synced across every assistant its members use. Write a note in Claude Code and ChatGPT knows it; connect Gmail once and Cursor can search it without a key sitting in a config file. The memory lives as plain markdown in a folder the team owns, so leaving takes the notes with you, and the open source parts are free to self-host.

Tools, skills, and a multiplayer workspace round it out, with scheduled jobs that run while you sleep and switchable Anthropic and OpenAI models. One price covers everyone invited. The launch offer is $99 for the first month, $199 after, including $100 in monthly usage credit, though that intro price expires September 29. The name-dropping in the positioning is a little much, but the ownership model is genuinely appealing.

Speechka

Speechka handles live translation without subtitles. You speak your language, others hear theirs, and the output arrives in your Personal voice after a ten second sample. Ordered playback keeps sentences in sequence when several people talk, and context-aware processing aims for natural phrasing rather than word-for-word output. Listen Mode plays translation to your headphones; Broadcast Mode pushes it through your mic so participants hear it directly.

It works with anything that accepts microphone input, including Zoom, Meet, Teams, Discord, and OBS. There is a 5-minute free browser trial and a 7-day desktop trial with 10 credits. Pro is $19.99 a month for 180 credits, and extra credits are sold separately. The credit model is the most confusing part of an otherwise clean pitch, and 44 languages is a solid coverage claim.

Quick links: