Almanac's YC S26 founders turned an internet-wide wiki experiment into an AI agent with plans starting at $30 per user per month across Slack, email, GitHub and the browser.
By RuntimeWire Staff · Published
Primary source: Almanac
Why it matters #
Almanac is betting that shared memory, rather than another chat interface, is the missing layer for workplace agents. Its early permissions gap shows how quickly that thesis becomes a security problem.
Rohan Sharma, Kushagra Chitkara and Divit Sheth launched Almanac, an AI workplace agent that compiles a team's scattered activity into a self-updating wiki, then uses that memory to carry out work through its own cloud computer.
The three founders met 11 years ago while preparing for India's JEE engineering entrance exam. Their paths later split across IIT Delhi, IIT Kharagpur, BITS Pilani, Harvard and engineering or research work at Arcesium, the Vector Institute, Carnegie Mellon and MIT. They reunited around an idea that had followed them through several products: an agent becomes useful when it retains context instead of starting each session with an empty prompt.
Almanac's public launch activity appeared during the week of August 27-30 as part of Y Combinator's Summer 2026 batch. The launch moves the founders from experiments in public knowledge toward a product built around private company information, where useful memory carries a clearer business case and much higher security stakes.
A wiki obsession gets a commercial target
Sharma studied electrical engineering at IIT Delhi before completing graduate work in computational science at Harvard. A Harvard profile described his research interests as theoretical deep learning and its use in climate and sustainability work. Chitkara studied electrical engineering at IIT Kharagpur and data science at Harvard. Sheth, Almanac's CTO, previously worked as a senior software engineer at financial technology provider Arcesium after studying electrical and electronics engineering at BITS Pilani.
The trio's earlier projects treated the wiki as an AI-native format rather than a set of pages employees must remember to maintain. In 2025, the founders released an AI biography product that generated sourced profiles from the public web. That effort expanded into a broader public knowledge base before the founders redirected the core idea toward workplace context.
In Almanac's YC launch note, the founders said they also spent roughly 18 months building agent wikis for Harvard and NASA and attempting to organize the wider internet. Those institutional projects remain company claims. The strategic progression is visible in what Almanac sells today: the wiki survived, while the scope narrowed from the internet to a single organization with recurring work and a budget.
"AI will need a context layer, and it takes the shape of a wiki," the founders wrote in the YC note. Almanac is their attempt to package that thesis as a daily operating system for small teams.
The product is memory plus execution
Almanac connects to services including Gmail, Google Calendar, GitHub, Slack, documents and meeting notes. It extracts decisions, customer details, project updates and unfinished tasks into personal and shared wikis. Each wiki entry links back to source material, and users can edit incorrect information so later runs use the correction.
The execution layer separates Almanac from a conventional knowledge base. Almanac says each agent has a browser, terminal, files and account logins running on an always-on cloud computer. A user can ask it through Slack or iMessage to review customer feedback, reproduce a reported bug, file a GitHub issue, prepare a meeting brief or handle work in a service without a dedicated integration.
The founders also want the wiki to supply context to other agents. Almanac says Claude Code and Codex can read the same organizational memory, reducing the repeated explanations that accompany new coding sessions. That pitch positions Almanac as shared infrastructure underneath several agents, rather than another isolated assistant with its own private history.
Automations are written as sentences instead of assembled in a workflow builder. Chitkara explained in public launch comments that recent information is incorporated into immediate context, while a nightly process reviews the day's sources, resolves contradictions and consolidates durable information into the wiki. Almanac calls that process "dreaming."
Almanac charges $30, $100 or $200 per user each month, depending on model usage and browser minutes, after a seven-day trial. Enterprise pricing is custom.
Permissions are the first test
Almanac's broad access is central to its usefulness and the largest obstacle to adoption. An agent expected to read email, monitor customer conversations, use financial accounts and operate a browser needs extensive credentials. Those same credentials expand the cost of a mistaken action, compromised session or faulty permission boundary.
Almanac says personal accounts remain restricted to the user who connected them, while organization-owned accounts can be made available to members according to workspace settings. The privacy policy says Google Workspace data is not used to train generalized AI models and that OAuth credentials are held by an account-connection provider rather than Almanac's product database. Almanac lists Browserbase for managed browser sessions, AgentWeb for credential storage, Supabase for product records and OpenAI for model inference on its subprocessor page.
The current permission model is still coarse. Asked whether a user could grant read-only email access, Chitkara wrote that Almanac currently requests all permissions upfront. After another prospective user raised the same concern, he said scoped access would be added to the roadmap.
That exchange matters because Almanac is asking customers to cross a trust boundary before Almanac has an established record inside large organizations. Human handoffs at logins, payments and consequential decisions constrain some actions, but they do not replace granular access controls. The founders' response shows an early product learning arriving in public: persistent company memory requires equally persistent permission discipline.
Selling into a funded category
Almanac is entering a market already occupied by enterprise search providers, knowledge-management products and agent platforms. Glean raised $150 million at a $7.2 billion valuation on June 10th, 2025, as it expanded from enterprise search into AI agents. Dust raised $40 million on May 18th, 2026, around a similar argument that agents need shared organizational context.
Almanac is approaching the problem from the smaller-company end of the market, with self-service pricing and an agent designed to act across tools immediately. Acceptance into YC carries the accelerator's standard $500,000 investment: $125,000 for 7% and $375,000 through an uncapped most-favored-nation SAFE.
The founders' wiki thesis gives Almanac a coherent technical and commercial direction after a year of broader knowledge experiments. The work ahead is less philosophical. Almanac must show that its memory stays accurate, its actions stay bounded and its permission controls can mature as quickly as the agent's reach.