Yesterday's Top Launches: 3 Tools from November 11, 2026
Three November 11 launches span a single-file PDF planner for iPad, a media search index that reads text from photos, and a memory layer for AI agents.
Three launches from November 11: a planner, a media search index and a memory layer for agents
Yesterday’s launch list was short and strangely mixed. Alongside the usual crop of new developer tools sat a 172-page PDF planner for iPad users and an AI index that reads the text on signs inside your photos. Three products, three completely different audiences, and only two of them would make sense in a terminal.
ClearWeek 2027
Start with the odd one out. ClearWeek is a digital planner shipped as a single landscape PDF, meant to be imported into GoodNotes, Notability or whatever annotation app you already use with a stylus. There is no app, no account and no sync service. The file is the product.
That constraint turns out to be the interesting part. Because the whole thing lives in one document, yearly planning, monthly resets and weekly reviews share the same pages instead of scattering across four apps. The structure runs from a Year Overview with US federal holidays marked, through Big 3 goals and four quarterly pages, into 12 monthly sections with calendars and habit trackers, then down to 53 weekly spreads. Each week opens with three priorities, has weekday focus blocks, and closes with a Friday shutdown checklist and a review. A Waiting On list handles the tedious business of chasing other people.
Practical notes: the dated calendar starts December 28, 2026 and runs to January 2, 2028, Monday start. The file is about 2.6 MB. Navigation depends on hyperlinks, and the product page admits those can behave differently depending on which annotation app you open it in, which is a fair warning rather than a marketing line. There are no daily pages and no sticker packs, so if you plan hour by hour this will feel sparse. Price is $14, discounted to $8 with code CLEARWEEK8.
Who it suits: someone who already thinks in weeks, owns an iPad and a stylus, and wants the year to stay visible while they work. If you plan on paper, this is not for you.
Searchunk
Camera archives are named badly, and everyone knows it. A photo of a veterinary clinic sign is called IMG_4471.HEIC, and six months later nobody can find it. Searchunk indexes what is actually visible in each image and each video scene, then lets you type a description like “dog running on the beach” and get ranked matches back with confidence scores.
Two details stand out. First, it reads text inside images, so signage, posters and screens become searchable, and it handles Turkish including suffixes, meaning a search for “kuşlar” also surfaces files described as “muhabbet kuşu”. Second, video is split scene by scene, so results carry timestamps and clicking one starts playback at 00:42 instead of the beginning. For anyone who has scrubbed through a forty-minute recording looking for four seconds of footage, that alone justifies a look.
The architecture is the selling point for teams with compliance people in the room. Your files never move. Searchunk connects to S3, Google Cloud Storage, Azure Blob, R2, Wasabi, Backblaze B2 and a long tail of S3-compatible providers, or to a local folder via a small agent. Only search data and downscaled previews are kept, and every model runs on Searchunk’s own servers rather than being handed to an outside AI vendor. Changed and deleted files are picked up automatically.
Worth flagging: it is in public beta, so outages are explicitly expected. PDFs and other documents are not indexed. Signup requires a company email, so freelancers on Gmail are locked out for now. The free tier covers the first 250 images and videos with every feature included, including teammate invites.
Willder
If you run more than one coding agent, you have already hit this. Claude Code learns a decision, Codex asks the same question an hour later, and you retype the answer. Willder is a context layer that sits between the agents and the team, so a fact saved once is available to the agents you shared it with, and only those.
The access model is the part I’d scrutinise. Each agent gets its own scoped, revocable token, and every read and write lands in an audit log. Revoke a token and the next call returns HTTP 401, logged. An agent on a marketing scope asking about a launch gets “No facts found.” Willder is upfront that revocation cannot un-recall information an agent already received, which is the honest version of that promise.
On consistency, facts are versioned and stale writes are refused, so two agents editing the same record do not silently clobber each other. The published test checked 30 conflict cases across 12 read paths and caught all of them. Deletion propagates to live stores within a minute, though backups are not yet covered.
The numbers are published with caveats attached, and the caveats matter. Six agents writing saw a new fact reach the others in a median of 538 ms. On the GateMem benchmark, five of 21 episodes scored 71.3 live, but private details leaked in 6.7% of privacy checks and deleted ones in 4.4% of deletion checks. The 19.8% token reduction is a projection from 48 replayed sessions, not a live measurement. Connectors cover GitHub, Google Drive, Slack, Figma and PostHog; Gmail, Notion, Linear and Jira are planned. Pricing is published but marked as not yet for sale. The engine is MIT-licensed and self-hostable, which is the strongest argument for trying it before the commercial plans open.
No community rankings had settled by the time of writing, so consider this a first look rather than a verdict.