Almanac: An AI Agent with a Self-Updating Company Wiki
Almanac (YC S26) is an always-on AI agent that maintains a self-updating wiki of your company's knowledge by connecting to your existing tools (Gmail, Slack, Google Calendar, PostHog, etc.). It runs on its own computer, so it can act proactively and follow up on tasks over days, not just single sessions.
How It Works
Almanac compiles information from your connected sources into two wikis: a personal wiki (understanding you, your preferences, and your life) and a company wiki (what the company is, what you're working on, roadmap, blockers). This pre-compiled knowledge layer allows the agent to “just know you.”
Key features:
- One-click connections to Gmail, Calendar, Granola, PostHog, and more. Personal accounts are only accessible by you; shared accounts are accessible by everyone.
- Wiki entries link to their source, so you can verify every claim. If the wiki is wrong, you can correct it and Almanac learns from the correction.
- Proactive agent: a background worker suggests tasks it can complete, e.g., “I already prepared a draft of your fundraising pitch deck, want to take a look?”
- Long-horizon tasks: Almanac keeps context on ongoing projects, so it can pick up a task days later, e.g., sending a follow-up email after noticing a reply.
- Own computer: Almanac uses its own browser, files, and logins, even for tools without an integration.
Privacy and Security
Almanac only accesses accounts you explicitly connect. OAuth credentials are held by the connection provider, not in Almanac's database. Only the wiki and the source behind citations (e.g., an email used as a citation may be retained as Markdown) are stored.
Who It's For
Teams that want an AI agent that actually gets tasks done across their existing tools, without having to re-explain context every time.
Almanac is live at usealmanac.com with a 7-day trial on all plans.
📖 Read the full source: HN AI Agents
👀 See Also

Why AI Coding Agents Churn Out Slop After 20 Turns: Context Blindness
A deep audit of API logs reveals Cursor and Claude Code aren't getting dumber — they suffocate on bloated context windows filled with noise, causing architectural destruction.

Codesight CLI reduces AI coding agent token usage by scanning codebases
Codesight is a zero-dependency CLI tool that scans TypeScript, Python, and Go projects to generate compact context files, reducing Claude Code exploration tokens by 12.3× on average according to benchmarks from real production codebases.

OpenIntel Iran: AI-Powered Conflict Dashboard Updates with Hourly Briefings
A developer's AI agent has overhauled the OpenIntel Iran dashboard, an auto-updating intelligence tool that scans Reuters, AP, BBC, and other major sources hourly to verify developments and publish structured briefings on the Iran-Israel-US conflict.
Needle: A 26M Parameter Function-Calling Model That Runs at 6000 tok/s on Mobile
Cactus open-sources Needle, a 26M parameter model for single-shot function calling, achieving 6000 tok/s prefill and 1200 tok/s decode on consumer devices. Built with Simple Attention Networks (no FFNs), it beats several larger models on tool-use benchmarks.