Almanac (YC S26) Review: What It Is and Who It's For
First, which Almanac
Two companies use this name and searching for one returns the other.
This review is about Almanac at usealmanac.com, a three-person Y Combinator Summer 2026 company in San Francisco that launched on Hacker News on August 31, 2026.
It is not Almanac Labs at almanac.io, the document-collaboration company founded in 2019 that raised a $9M seed led by Floodgate and a $34M Series A led by Tiger Global. That company is older, larger, also sells a docs-and-wiki product, and owns the search results for the word. If you arrived here looking for the one with the Series A, you’re in the wrong place.
We read the prices and product details below off the company’s own pages and its Launch HN thread on September 7, 2026.
What Almanac is
An agent with its own cloud computer. It has a browser, files, a terminal and its own logins, and it compiles the accounts you connect into two self-updating wikis: a personal one and one for the company. Their own line for it is that work happens in your tools, Almanac compiles it into a wiki, then reads that before doing anything.
You reach it by text, in Slack or iMessage, on every tier. The tools it names include Slack, Gmail, Google Calendar, Google Docs, Chrome, GitHub, Granola, plus Mercury, Vercel, Notion, Linear, PostHog and Attio. A background worker proposes tasks it could take on, and the wiki’s project section is what lets one of those resume days later.
The thesis is that an agent is only as good as what it knows about your company, so the knowledge layer should assemble itself rather than being written by hand. We think that thesis is right. It’s the one we build on.
What the founder said in his own thread
The Launch HN post drew 59 points and 48 comments, and the founder answered four questions that narrow the product considerably.
The wiki updates once a day. Asked what happens when someone deletes a source email or revokes access, he said the fact disappears on the same daily pass. So “self-updating” means a daily batch compile rather than something that reacts as things change.
Permissions are binary. Content is personal to you or shared with everyone in the company. Two separate commenters pushed for something in between, one asking for me, my team, my department, whole company. He moved granular permissions up in priority during the thread.
One model, no choice. GPT-5.6-terra under the hood, with multi-model and bring-your-own-key on the roadmap. A CLI exists that exposes the wiki layer to another harness.
No benchmarks. Asked for evidence that the memory approach improves performance, he said they have no formal benchmarks, and explained why he thinks existing memory benchmarks do not measure what matters here.
Also from the thread: the browser runs on Browserbase, two-factor prompts hand control back to a human mid-login, and he acknowledged that sites which ban bot traffic can suspend the account.
What it costs, and what the tiers actually buy
| Plan | Price | What changes |
|---|---|---|
| Starter | $30 / user / month | Standard model usage, standard browser minutes |
| Pro (most popular) | $100 / user / month | 5x model usage, 2x browser minutes |
| Scale | $200 / user / month | 20x model usage, 4x browser minutes |
| Enterprise | Custom | SSO, access controls, rollout support |
Everything else is the same on every tier: unlimited connected accounts, the shared company wiki, Almanac for teammates and shared channels, Slack and iMessage.
Read the middle column again, because it’s the whole pricing model. The tiers are graded purely by model usage and browser minutes. That’s a credit ladder wearing seat pricing, so the list price is where a customer starts rather than where they land. A team doing light work pays $30 a head. A team running the agent hard moves up the ladder.
It is a waitlist today
The pricing page says Almanac is in alpha, every button reads JOIN WAITLIST, and there is no annual option and no trial anywhere on it. The Launch HN post a week earlier said they were live with a seven-day trial on all plans.
Those two things disagree, and the page is the more recent. As of September 7, 2026, what Almanac offers is a place in a queue. Treat any review claiming hands-on experience of the shipped product, including the pricing above, as describing something you cannot buy yet.
What the Hacker News thread actually said
Reception was lukewarm, and the substance is more useful than the score. Fifty-nine points is modest for a Launch HN, and the recurring theme was that nobody could tell what made it different. One commenter wrote that the website told them nothing about what makes this new or unique. Another listed roughly twenty-five rivals by name and asked how Almanac differs; the founder answered with the wiki layer and no numbers, and a third commenter said that was not compelling and asked for something concrete.
Two separate commenters named Microsoft’s WorkIQ as the same promise, and one said it looked identical to Claude Desktop connectors. That’s the reflex comparison a technical audience now reaches for whenever a startup says it’ll know your company, and it’s the objection this category has to answer.
Security
OAuth credentials sit with their connection provider rather than in Almanac’s own database. They store the wiki and the sources behind its citations, which they describe as meaning an email used as a citation may be retained as Markdown. The footer links a privacy policy, a subprocessor list and terms.
No SOC 2 or any other certification is claimed anywhere we could find. At three people in alpha that’s unsurprising, but if a security review is a gate you have to clear this quarter, ask about certification before anything else.
Where we fit
We build this+that, so read this section knowing that.
We agree with Almanac’s premise. A company’s knowledge should assemble itself out of the work rather than being written down by somebody who remembers to. That’s what the Brain is for, and it’s why this company is worth watching at three people and a waitlist.
Two differences are worth naming.
What starts the work. Almanac acts when you text it, or when its daily pass proposes something. We start when somebody else’s message arrives. A daily compile is a clock rather than an event, and a clock can’t help with the request that landed twenty minutes ago.
What gets read. Almanac’s sources are mostly tools of record: GitHub, Attio, Notion, Linear, PostHog, alongside Gmail and Calendar. We read the channel stream, which is where things arrive before anyone files them anywhere. A tool of record only holds what somebody already decided to write down, and the gap between what a company knows and what its systems hold is mostly things nobody filed.
Our own workflows write back to the Brain as they run, and zero-configuration upkeep is still ahead of us. So the contrast is daily batch compilation against write-back driven by workflows, not stale against live.
Frequently asked questions
Is Almanac available to buy?
Not as of September 7, 2026. The pricing page describes an alpha and every call to action joins a waitlist, even though the Launch HN post a week earlier said they were live with a trial.
How often does the Almanac wiki update?
Once a day. The founder confirmed in the Launch HN thread that updates happen on a daily pass, and that removals propagate on the same schedule.
Can I control who sees what?
Not granularly today. Permissions are personal or company-wide, and the founder said in the thread that finer controls moved up the roadmap after two commenters raised it.
Which model does Almanac use?
GPT-5.6-terra, with no choice of model. Multi-model support and bring-your-own-key are on the roadmap, and a CLI exposes the wiki layer to another harness.
Does Almanac have SOC 2?
No certification is claimed on the site or in the launch post.