Show HN: Myna – a local AI Chief of Staff that remembers your work Myna, a local AI chief of staff built on Claude Code, launches to help tech professionals manage projects, meetings, email, and team context entirely on-device. The system uses 31 skills and a persistent local knowledge base to prep briefs, triage inboxes, and draft updates without sending data to external servers, aiming to reclaim the 80% of workweek spent on information management. The system for everything you're trying to keep in your head. Nothing leaves your machine. AI Chief of Staff for tech professionals — your projects, your team, your meetings, your email, all running locally inside Claude Code https://claude.ai/code . Built autonomously by Claude Code, end to end — 12 agents, 31 skills, no vibe-coding. How it was built → It's 7:45 AM. You haven't opened your laptop, and you already know roughly what's waiting. A 9 AM 1:1 with someone who's been stretched thin — and you still haven't written down the pattern. Six meetings stacked back to back, none of them prepped. An inbox of 38 unread, four of which are real and the rest of which look real. A VP who wants a "quick Phoenix risk note" by Friday, which means remembering what happened in week one of Phoenix — 47 days ago. Two people owe you something; you owe four people something back. And somewhere in there, you're supposed to do your own work. You're not behind. This is just the job. If you manage projects and people, most of your day isn't decision-making — it's information management: triage, prep, catch-up, drafting the same kind of update for the fifth time this month, trying to recall what was decided three weeks ago. Roughly 80% of the week is moving information around. The other 20% is the work only you can do — and it's the part you actually get paid for. The tools you already have don't fit the shape of that 80%: General AI chat Claude, ChatGPT, Gemini starts every session from zero. It doesn't know who Marcus is, what Phoenix is blocked on, or how you sign your emails — so you spend the first three minutes re-explaining context, and none of it survives the closed tab. Local AI notebooks Khoj, Fabric, PrivateGPT answer questions about notes you already wrote. They don't do the work — no prep brief, no triage routing, no drafted escalation. They're search engines with a chat box, not assistants. Cloud "AI assistants" for work want to live in the middle of your stack and ingest your email and calendar onto someone else's servers. At most companies that's a non-starter — and even where it's allowed, it's one more vendor to evaluate and one more thing to break. Myna sits exactly in that gap. Myna is a Claude Code https://claude.ai/code agent with 31 skills for the work that fills your week: it preps your meetings, triages your inbox, keeps every project and person up to date, and drafts your replies and status updates — all from prompts you type in plain language. It connects to the tools you already use through MCP, reading from your email, Slack, and calendar, and maintains a local knowledge base of your work as plain markdown: projects, people, meetings, decisions, action items, and history. Plenty of tools can draft an email or summarize a thread. What sets Myna apart is the context behind every one: it works from that knowledge base instead of a blank slate. Install it once and every session builds on the last — the brief on Sarah this morning already includes the recognition you logged last week and the design review she led on April 9, and the Phoenix risk note you draft today knows the blocker is 20 days old, because Myna wrote it all down when it happened. Its drafts sound like you and reflect what's actually going on. It drafts but never sends. It organizes but never decides. It surfaces but never hides. You still pick up the phone, send the email, and make the call — Myna just makes sure that when you do, you're not winging it from memory. A day in the life of an engineering manager using Myna. Every line is a prompt you'd actually type into Claude Code. 7:45 AM — coffee, laptop open daily-brief ↳ Daily note created. Today's meetings as a table with linked prep. Phoenix blocker flagged. Sarah Mitchell's reply overdue. 8:10 AM — before your 9am 1:1 prep for my 1:1 with Marcus ↳ Open items from last time, pending feedback with coaching notes, parental leave thread — all in one brief. 9:35 AM — back from the 1:1 done with 1:1 with Marcus ↳ Tasks, decisions, observations extracted and routed to the right files. 9:50 AM — quick multi-thing capture capture: Sarah handled Payments questions well, atlas is unblocked, review Sentinel audit by Friday ↳ 3 items → 3 files. Recognition, timeline update, task with due date. 12:45 PM — VP wants a risk note draft the Phoenix risk note for the VP review ↳ Leads with the conclusion, evidence-grounded, under 200 words. Ready for you to review, then send yourself. 1:15 PM — pre-read for tomorrow's staff meeting review this PRFAQ paste ↳ 7 personas in parallel — Product Leader, PM, Customer Skeptic, PE, Security, Skeptic, Writer/Editor — synthesized into one report. Saved to Reviews/2026-05-26-pricing-prfaq.md. 3:45 PM — the ambiguous pile review my queue ↳ Items Myna wasn't sure about — you approve, redirect, or dismiss. 5:30 PM — close the day wrap up ↳ Planned vs actual. Contributions logged. Tomorrow's note created with carry-forwards. Not an EM? The same skills work for tech leads, senior engineers, and PMs — anyone managing projects, people, and a constant stream of communication. Full walkthrough: A Day With Myna /agentflock/myna/blob/main/docs/guide/a-day-with-myna.md · Browse the demo vault /agentflock/myna/blob/main/tests/fixtures/vault/myna to see the files Myna creates. Install Myna inside Claude Code in three lines: /plugin marketplace add agentflock/plugins /plugin install myna@agentflock /myna:setup You'll be running your first prompt in about five minutes. Prerequisites and run modes are in Getting started getting-started below. Notice what those prompts have in common: not one needed a format, a tone, or a destination explained. Three things make that possible — and a fourth makes Myna yours to reshape. 1. It's purpose-built for the work you actually do. Myna isn't a generic "AI assistant" prompt — it's 31 task-specific skills, each built for a real moment in a tech professional's week. Each one already knows what to read, what to write, where the result lives, and what to refuse. You don't describe the shape of a 1:1 prep or a VP risk note; the skill has done this job before. That's the difference between a chat box and an assistant. 2. It builds a knowledge base that's yours, on your machine. Every session writes back to your vault, and the next session reads it. So "brief me on Marcus" pulls six months of 1:1 history, observations, and pending feedback you logged without thinking about it; "draft recognition for Sarah" cites the actual April 9 design review; "what did I ship this quarter?" becomes a real query instead of a vibe. It's all plain markdown in one folder — readable, editable, greppable, and yours to walk away with the day you stop using Myna. 3. It's safe for the enterprise you actually work at. Myna writes only to that one local folder, drafts instead of sending, and adds zero new infrastructure — it reuses the MCP servers your security team already approved. There's nothing new for them to evaluate. The privacy section below spells out exactly how each guarantee holds. 4. It's yours to keep changing — with the same harness that built it. Myna doesn't make you wait on a roadmap. It ships the dev skills Claude used to build it, so when you want a new skill, hit a bug, or need an existing skill to behave differently, you describe the change in plain language and a guided design → build → review → PR loop makes it real. Most assistants let you flip a few settings; Myna hands you the toolchain that builds it. Here's how that lands against the two kinds of tools you've probably already reached for: AI Chat Tools Claude, ChatGPT, Gemini | Local AI Tools Khoj, Fabric, PrivateGPT | Myna | | |---|---|---|---| Built for | General-purpose chat | Q&A over your notes | Your workday — daily brief, meeting prep, email triage, project catch-up, doc review | Your data | On a server, general-purpose memory | Your existing notes, unstructured | Your projects, people, and meetings — organized as you work | What it creates | Chat responses and one-off documents | Answers about your notes | Drafts, prep, briefs, daily notes — files you review before using | How it works | You direct every step | One question, one answer | One prompt — Myna routes, updates, and files across your vault | Integrations | Whatever you describe | Your local files | Email, Slack, and calendar via your company's existing MCPs | Customizable | Settings and custom GPTs | Config file or source code | Plain-text skills you can read and edit — or reshape with the bundled dev harness | 31 skills, grouped by the part of your day they belong to. You don't memorize them — you type natural language and Myna routes to the right one. Each example below is a real prompt. Your daily rhythm. Start the day, capture in passing, close it out. daily-brief — your morning note: today's meetings as a table with linked prep, overdue tasks, vault reminders surfaced, review queue flagged what should I focus on today? — ranked priorities, grounded in actual blockers and unreplied threads capture: Sarah crushed the Payments review, atlas is unblocked, remind me Friday to check the Sentinel audit — one sentence routed to three files at once remind me on Friday to review the deploy checklist — vault reminder written immediately, surfaces in your daily brief, with an optional calendar push if you give a time park this / resume auth caching — zero-loss context switching across sessions wrap up — planned vs actual, contributions logged, tomorrow's note seeded Your inbox and Slack. Cut the noise, route the signal. triage these inbox emails — folder recommendation and one-line reasoning per email; anything needing your reply is flagged Action Required , with an overdue marker once the deadline has passed process updates — reads your project-mapped Slack channels and email folders and routes the structured data into your vault: project updates, action items, and timeline entries land in the right files process instructions — reads a configured folder for the instructions you leave Myna while clearing your inbox, then carries them out — draft a reply, add a TODO, and so on Your meetings. Prep before, process after. prep for my 1:1 with Marcus — open items, pending feedback with coaching framing, recent context prep for my remaining meetings today — skips the already-prepped, fills in the rest done with 1:1 with Marcus — tasks, decisions, and observations extracted and routed in one step reserve 2 hours Monday for the coverage plan — personal time blocks only; Myna never creates events with attendees Your projects. Stay current without digging. catch me up on atlas migration — timeline, blockers, tasks, dependencies, upcoming meetings what's blocked? — every blocker across every project, with age and next action what am I waiting on? — what needs your reply vs. what you're owed break down the "review Sarah's caching design doc" task — subtasks with reasoning Your people. Briefs, patterns, and the gaps you can't see. brief me on Sarah Carter — role, shared projects, open items, pending feedback, 1:1 history how is my team doing? — portfolio view: tasks, overdue, feedback gaps, last 1:1 per person analyze my 1:1s with Marcus — patterns, follow-through rate, recurring topics who haven't I given feedback to in a while? — feedback-gap alerts sorted by staleness build Sarah's performance review narrative — synthesized from months of observations and recognition Your writing. Grounded in context, in your voice. draft an escalation for the Phoenix validator blocker — severity, impact, recommended action, grounded in project history draft recognition for Sarah Carter for the design review this week — specific and evidence-backed, never generic fix this: i wanted to loop you in on sarahs progress... — grammar and tone cleaned up, your voice preserved draft reply to James about the Q3 staffing proposal — pulls from James's person file and related projects Your own track record. Stop underselling yourself. log contribution: led atlas design review, got cross-team alignment — categorized and appended to this week's log build my brag doc for Q1 — assembled from your contributions, organized by impact am I underselling myself in this self-review? — checks your draft against what you actually logged Documents that hit your desk. Multi-persona review of any technical doc. review this PRFAQ: paste — Product Leader, PM, Customer Skeptic, PE, Security, Skeptic, and Writer/Editor run in parallel, synthesized into one report review this design doc: paste — composes the right panel PE, Sr SDE, SRE, Security, QA for the doc type- 11 personas available in total; reports save to Reviews/ And underneath all of it, the review queue. Anything Myna isn't sure about — an ambiguous owner, an inferred contribution, an unclear intent — lands in a persistent queue instead of resolving silently. review my queue walks you through them one at a time, and you approve, redirect, or dismiss. The User Guide /agentflock/myna/blob/main/docs/guide/guide.md covers all 31 skills in the order you'd use them, with what each one reads and writes. Myna was designed for people whose company won't let them paste a customer email into ChatGPT. The guarantees are specific, layered, and verifiable — and they're the reason the differentiation above is more than a marketing claim. All writes are local. Every Myna write lands inside one folder,