Unabyss
MCP-native self-updating context layer for your AI
1.8K followers
MCP-native self-updating context layer for your AI
1.8K followers
Set it up once and never re-explain yourself to AI again. Connect the apps you use daily - Unabyss will extract, structure, and update your context automatically. Share it with any AI tool via MCP, with granular control over what each tool can see.
Unabyss
Hey PH 👋 Philip here, co-founder of Unabyss.
What is Unabyss? Unabyss is your personal context layer - a single, structured vault of your identity, knowledge, and preferences that any AI app or agent can access instantly, with you in full control of what gets shared and with whom.
The Problem Every AI tool you use starts from zero. You re-explain your role, your goals, your tone, your company - over and over. And when you finally do build up context inside one platform, it's trapped there. ChatGPT memory doesn't follow you to Claude. Claude Projects don't talk to Cursor. The more AI tools you adopt, the worse it gets.
The Solution Unabyss extracts your context once - from LinkedIn, your website, Notion, Gmail, Slack, GitHub, and more - and structures it into clean, layered files (persona.md, voice.md, company.md...). From that point on, every agent and LLM tool you use can pull exactly the right context automatically, via MCP or one-click exports. No re-explaining. No copy-pasting. No context left behind.
What makes it different: your context is user-owned, pre-extracted (not built from interactions over time), and cross-platform - it works with any tool, any LLM, any agent, through a single connection.
Key Features
⚡ Auto-extraction from your existing tools in under 90 seconds
🔒 Granular permissions — share e.g. voice.md without exposing professional.md. iOS-style control, not cookie banners
🔌 MCP server for Claude, Cursor, Claude Code, OpenClaw, and any compatible agent
📤 One-click exports - investor updates, meeting prep, ICPs, bios - generated from your context instantly
🔄 Always up to date as your sources sync
Who It's For Founders, operators, and builders who live across multiple AI tools and are tired of starting from zero every time. If you use Claude, Cursor, ChatGPT, or any LLM daily - and you've ever thought "it should already know this" - Unabyss is for you.
What We'd Love From You Try it, connect your first source, and tell us: which integrations should we prioritize next, and where do you need your context the most? We'll be here all day reading every comment — your feedback directly shapes what we build next. 🙏
@philip_kubinski Hey Philip, congrats on shipping 👋
The "pre-extracted, user-owned, cross-platform" positioning is sharp, and pulling context out of interactions into clean layered files (persona.md, voice.md) is the right structure most memory tools miss.
One question on the freshness claim. You say context stays "always up to date as your sources sync," but pre-extracted context has the same failure mode as any cache: it drifts. If Unabyss pulled my identity and preferences from LinkedIn and Notion three months ago, and since then I changed roles, repositioned my company, shifted how I talk about what I do, what actually triggers a re-extraction? Source-change detection, a refresh schedule, or me manually telling it "this is stale now"? Asking because the gap between "structured once" and "actually current" is exactly where context layers quietly start lying to the model with confidence.
Unabyss
@philip_kubinski @arturbrugeman
Hi Artur! Context staleness is indeed one of the most seamless-feel-breaking issues across similar solutions! We address this primarily by refreshing your social data at least once per day (unless you opt-out) and comparing to that to what we've seen before. Any really major shifts are immediately noted not only in our vector search DB but also injected into prompts at multiple call-sites, to make sure all queries are accurate and up-to-date.
So to answer your question
> [...] what actually triggers a re-extraction? Source-change detection, a refresh schedule, or me manually telling it "this is stale now"?
We use all three ✅ ✅ ✅
@philip_kubinski @malpunek "Refresh daily, compare to last seen, inject major shifts at all call-sites" is a solid answer, and the daily cadence is the right default for social-sourced context where things actually do change week to week.
The one I'd watch long-term is the silent-drift case: not a role change Unabyss can detect from a LinkedIn diff, but a slow shift in how someone positions themselves that never shows up as a hard signal in any source. That's the hard 10% no refresh schedule catches. But that's a frontier problem, not a launch-day one.
Good answer, thanks Stanisław.
Unabyss
@philip_kubinski @arturbrugeman
We're tackling the last 10% with a variety of heuristics, these should shrink the 10% even further. AFAIK there's no one canonical way to address this perfectly today. If you have some further thoughts and ideas about it I'd be happy to chat!
bunny.net
@philip_kubinski congrats on the launch!
Unabyss
@marek_nalikowski thanks a lot, Marek! happy to hear your feedback once you test it :)
PicWish
@philip_kubinski How does the token efficiency and latency? compare when feeding Unabyss via mcp vs just using .cursorrules or claude projects?
Unabyss
Hi @mohsinproduct ! Token efficiency is much better since our MCP loads only the relevant parts; and the gap is wider the bigger your knowledgebase. For latency we're slightly worse - as with all MCP tool calls - but once your rules start to eat up a significant part of your context window Unabyss becomes the obvious choice
Unabyss
Okay, European shift is over, now we begin the Australian/Asian shift ;) Still fresh, still happy to answer any of your questions & hear your feedback on Unabyss!
@philip_kubinski This is a solid take the context fragmentation problem is real and only getting worse as people stack more AI tools.
What I find interesting though is that most solutions in this space are focusing heavily on memory and identity, but not really on what happens after that context is available.
In real workflows, especially on the agency/operator side, the bottleneck usually isn’t “what does the AI know about me” it’s “what do I actually do with this to get outcomes like leads, clients, or revenue.”
That’s the gap I’ve been exploring with my own web app I just built more on the execution side of things, where context needs to turn into structured outreach and client acquisition workflows, not just storage.
Curious how you’re thinking about that downstream layer.
Unabyss
thanks@veerhunt_agai! About the scenario: when you sign up, Unabyss will create an identity summary of you and will ask you if there are any conflicts - once set there it will be the base source of truth about you. Down the road, when you use it in agents or in our context chat, if our segmentation engine can't figure something out, it will ask you.
How does updating work in practice? If my tone or role changes over time, do I manually refresh it or does Unabyss adjust it automatically from new activity?
Unabyss
@marcel_dybalski1 Unabyss will update it for you when something changes, so there’s no need to refresh it manually. It’s works like a self-updating memory :)
Unabyss
@lakshminath_dondeti the main challenge is to "remember" what's truly important and sift out the rest. And that's what Unabyss is best at. We structure, tag, and keep track of the changes so what's important is always available - no more confusion in your Claude.
Curious how context staleness is handled — if I update a doc in a connected app mid-conversation, does the MCP layer reflect that in real-time or on a sync interval?
Unabyss
@hirogure great question! you can set up the sync interval for each app connected to Unabyss (see the screenshot).
In the situation you described, context might be missing the update, but after 5min, it'll be imported & added to the vault.
do you see specific use-cases where the real-time sync would be essential?
Unabyss
@hirogure syncs with the apps, happens on schedule, so if you have set sync every hour, we will pull the changes then, not when you change something in the file.
Love the idea of a centralized context vault! Repeating the same brand guidelines and operational context to different AI tools is a massive friction point. Quick technical question: when sharing data via MCP, how does Unabyss filter what's actually relevant to the user's current prompt so it doesn't just nuke the LLM's context window and skyrocket token costs?
Unabyss
@andika_fadhilah great question, Andika! There's specific agent that decides what kind of information can be useful for this specific prompt :) So we don't nuke the context window - it's a precise & accurate context ingest
Unabyss
@andika_fadhilah thanks for the kind words ;). In terms of MCP, your agent, where the Unabyss MCP is connected, is deciding what to ask and retrieve from Unabyss. But on our end, we are optimizing the output to be as token frugal as possible.
Do you plan to be HIPAA and SOC 2 compliant in the future? I work in health tech and would love to use this, but we have high security standards we have to meet for new tools.
Unabyss
@toni_toomey1 Yes! SOC 2 is on our roadmap very soon, but HIPAA is probably a bit of a long term :/
Unabyss
@toni_toomey1 we are currently in the process of getting SOC 2 - others certifications will follow.