# Nessie Labs Nessie is a desktop app that collects every AI conversation and agent session you have, across tools, into one library. Then it hands that history back to you, your teammates and your agents. Their tagline: "All your AI work, in one place." The company behind it is Nessie Labs, Inc (nessielabs.com), a [[Y Combinator]] Fall 2025 startup based in San Francisco. Watch the name. Nessie Labs has NOTHING to do with Ness Labs (nesslabs.com), the learning community and newsletter of [[Anne-Laure Le Cunff]]. Different founders, different product. One letter apart. ## Why it exists The problem is real, and if you use several AI tools every day you already feel it. Your thinking now happens in [[ChatGPT]], [[Claude]], [[Gemini]], [[Claude Code]], [[OpenAI Codex]]... and every session dies when it ends. Each provider has its own memory, but that memory stays locked inside one app. The next session starts from zero, and so does the next tool. The origin story makes the point well. CEO Anna Y. Zhang had been using a single Gemini conversation as her main thinking space for months. Then it hit the one-million-token limit, and months of research became unreachable overnight. Their bet: the "trace" (the full record of what was tried, discarded, decided and why) is a new kind of data asset. Companies used to keep the final document and lose the ten conversations that shaped it. With AI, those conversations are written down anyway; Nessie wants to store them and mine them. Their vision essay calls the end state "the recursive company": one that learns from its own record. ## What it does - **Ingestion**: connect a source once, Nessie backfills all available history and keeps syncing. No copy-pasting, no Chrome extension. Sources include ChatGPT (with Projects), Claude, Claude Cowork, Claude Code (sessions and memory), Codex, OpenCode, [[Cursor]], Gemini, Grok, Perplexity, Manus, Pi, [[Hermes Agent]], [[OpenClaw]], [[Obsidian]] vaults, Granola, Read AI, Apple Notes, and Microsoft Outlook, Teams and SharePoint - **Search and chat** across that whole history - **Contexts**: structured Markdown documents (briefs, decision summaries, handoffs) generated from your past conversations. Each context links back to its source conversations, and you can edit them, collaborate on them live, or copy them into any AI tool - **Profile**: a living profile built from your conversations, in 10 sections (biography, work, expertise, decisions, projects, connections...). Four of them update automatically as new conversations come in - **Agent access**: a CLI, an [[Model Context Protocol (MCP)|MCP]] server, packaged skills for Claude Code and Codex, and plugins for OpenClaw, Hermes and [[Grok Bot]]. Agents can search your history, read full transcripts and resume an earlier session (e.g., pick up in Codex where a Claude Code session stopped). An optional hook injects a briefing at the start of each new Claude Code or Codex session - **Skills management**: create, share and install agent skills across agents and teammates, with usage analytics and an eval on every invocation - **Sharing**: share one session, a whole repository, an integration, a context or a folder with a person or a team, or through a public read-only link - **Secret redaction**: API keys, tokens, JWTs, database URLs and private keys get replaced with `[REDACTED]` before indexing ## Privacy model It's local-first by default: your library lives on your Mac, and provider credentials never leave the machine. Cloud Sync is optional, but MCP, teams, skills and multi-device sync all need it (MCP clients read the cloud copy). "Privacy Mode" is the default and means your content is never used to improve their systems. Summaries and search still go through third-party AI providers at request time, though they state that they don't store that content or train on it. ## Platforms - **macOS**: the main app - **Windows** (10/11, x64 and ARM64) and **Linux** (`.deb` packages, x64 and ARM64): Early Access, with Cloud Sync forced on - **Headless**: the `nessie-daemon` CLI (`npm install -g @nessielabs/daemon`) pushes Claude Code, Codex, OpenCode and Hermes sessions from servers, VMs and containers - **On-prem** deployment for organizations Sign-in is Google only. It's closed source (the npm package is `UNLICENSED`). ## Pricing Every plan starts with a 7-day free trial. Prices are billed yearly (as of October 2026): - **Plus**: $28/month. Up to 5 connected accounts, unlimited sync, MCP, CLI, sharing, with usage limits on context generation and chat - **Pro**: $79/month. Unlimited accounts, context generation and chat - **Team**: $80/seat/month. Adds an admin dashboard, team analytics, a shared skill layer, "Ontology" (a structured model of team context) and a dedicated Slack channel - **On-prem**: custom pricing ## Who's behind it Two co-founders, both Yale 2024 graduates and former Amazon engineers: - **Anna Y. Zhang** (CEO): built ML-powered features at Amazon's recommendation service. Did neuroscience research at The Rockefeller University in high school - **Tiger Wang** (CTO): built authentication systems at Amazon handling 40 billion requests per day. Researched Linux kernel security at Yale The YC profile lists a team of three, backed by BoxGroup, Pioneer Fund, Y Combinator and Precursor. They rebuilt the product three times in six months. First came "Infinite Chat", a premium chat with full memory of imported conversations (built in their first two weeks at YC). It died because most AI chats are short and random. Next came an auto-organizer that summarized and filed conversations into notes. That one launched in November with ~1,900 signups and 300,000+ conversations imported in a week, but it hit the "then what?" wall: an organized pile of chats you don't use. The current version dropped the word "notes" entirely and replaced it with "contexts": outputs you generate, share and deploy. Since then the focus has moved from individuals toward teams and agent traces. ## How it relates to PKM and other AI tools This is [[Personal Knowledge Management (PKM)]] applied to a new raw material. The capture source is no longer what you read; it's what you think through with AI. And contexts are basically generated notes that come with provenance. The design philosophy goes the other way from classic PKM, though. In their words: "We don't offload taste to the user. Organization is our job, not yours." Most PKM tools hand you building blocks and let you design the system. Nessie decides what gets structured and what's stale. If you love owning your structure (I do), that's the tradeoff to weigh. If you never wanted to build a system in the first place, it's the selling point. It complements a vault more than it replaces one. Obsidian is one of its sources: Nessie indexes your Markdown notes next to your AI sessions and lets agents search both. The pitch is close to [[Context Engineering]] at the personal and team level; their blog calls the practice "contextmaxxing" (keep the full record, retrieve the few sources that answer the question, let the next agent continue from the current state). Alternatives and neighbors: - **Built-in memory** of ChatGPT, Claude and others: free, but it stays inside one provider, which is exactly the gap Nessie targets - **DIY**: export your conversations into an Obsidian vault and expose it to agents over MCP. You get full ownership, but you do all the plumbing yourself, and there's no live sync of agent sessions - **Other "company memory" startups**: when Almanac (YC S26) launched on Hacker News in August 2026, a commenter listed Nessie as a competitor alongside Mem0, screenpipe, Onyx, Slite, Dashworks, Hyperspell and others One gotcha worth knowing: Claude Code only keeps about 30 days of session history on disk, so your first import only goes back that far. Nessie keeps everything from the day you connect it. ## References - Homepage: https://nessielabs.com/ - About (origin, beliefs, team): https://nessielabs.com/about - Vision, "The Recursive Company": https://nessielabs.com/vision - Pricing: https://nessielabs.com/pricing - Docs (integrations, Obsidian, contexts, profile, MCP, Cloud Sync, privacy modes, data storage, secret redaction, daemon, Windows, Linux, Team plan): https://nessielabs.com/docs - Changelog (macOS v1.4.4656, 2026-09-29; Windows + Linux v1.1.1064, 2026-09-10): https://nessielabs.com/changelog - Privacy Policy (effective 2026-09-03): https://nessielabs.com/privacy - Blog, "Why We Killed Our Product Twice" (Anna Y. Zhang, 2026-04-19): https://nessielabs.com/blog/why-we-killed-our-product-twice - Blog, "Make your thinking legible to every AI you use" (Anna Y. Zhang, 2026-05-20): https://nessielabs.com/blog/make-your-thinking-legible - Blog, "Contextmaxxing Is Not About a Bigger Context Window" (Anna Y. Zhang, 2026-07-29): https://nessielabs.com/blog/contextmaxxing-not-bigger-context-window - Y Combinator company profile (Fall 2025 batch, founders, investors): https://www.ycombinator.com/companies/nessie - npm package `@nessielabs/daemon`: https://www.npmjs.com/package/@nessielabs/daemon - Hacker News, competitor list in the Almanac Launch HN thread (2026-08-31): https://news.ycombinator.com/item?id=49512486 - X: https://x.com/NessieLabs ## Related - [[Context Engineering]] - [[Personal Context Management (PCM)]] - [[Team Context Management (TCM)]] - [[Personal Knowledge Management (PKM)]] - [[Model Context Protocol (MCP)]] - [[Claude Code]] - [[Obsidian]] - [[AI Agents]]