Launched this week
Unabyss
MCP-native self-updating context layer for your AI
1.6K followers
MCP-native self-updating context layer for your AI
1.6K 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.
Can I have like a "work mode" and a "personal mode" and flip between them depending on which tool I'm using?
Unabyss
@chris_zawisza you can do it via permission-level and generate two MCP tokens:
- one enabling only personal context
- the other enabling professional context
We're also working on adding multiple email accounts at the same time - this should be ready and launched still this week.
Unabyss
@chris_zawisza @marcin_uchacz1 Love it when users ask questions about features that are next on our roadmap - thanks Chris!
@marcin_uchacz1 perfect, thanks for the answer!
@marcin_uchacz1 does this mean we'll be able to segment across jobs? I have multiple companies and one of the biggest challenges I have is keeping context appropriately segmented across tools.
viktor.com
how is this different from connecting your github repo (that has all the context about you)?
Unabyss
@toni_olendzki thx for asking! it depends what you mean:
1) repo as a personal wiki (Karpathy’s idea): although we really admire Karpathy's job, in his solution the AI writes and updates markdown notes from stuff you add. Works great for focused research, but not really for multiple data sources that need processing (nor from precision, nor retrieval efficiency perspective) + it takes time and you need to be fairly technical. Lastly, there’s no real permission level for each app / agent.
2) repo as the actual project code: even then, the repo is only part of the picture that LLM should see. A lot of how your team works lives in Slack, Linear, email, Drive, calls, and so on (not only in git)
Unabyss pulls all sources together, keeps them up to date, and enables AI models and agents to use your context safely and efficiently (saving lots of tokens)
SaveMyTime 3.0
Is there a risk that the Unabyss fills the gaps in my context with fabricated information? I mean - hallucination?
Unabyss
@adam_janczewski1 we have guardrails to prevent that, even 3 to be precise :D. But, as with any LLM tool, we can't say it's 100% preventable. We are constantly improving it, so it will be even better each time you use it.
Great idea - not my use case yet as I use only one AI tool and also have very well written personal/projects skills - but definitely will revisit this when I hit some issues with remembering me as person by AI in multpile AI tools!
Unabyss
@wojciech_sado you can take advantage of the Unabyss now. I would even say that your case is perfect. Your skills can be enriched with data from Gmail, Notion, or Obsidian, so they don't stay static as the input structure changes. Another use case is to run an agent to create new skills based on your carefully crafted ones for new input types.
persona.md/voice.md/company.md as plaintext files you actually own is the part that matters here. same reason a CLAUDE.md you can read and edit beats whatever ChatGPT stashes in its memory drawer. and being MCP-native
Unabyss
@webappski the real magic happens when these files are self-updating whenever you do something in Slack/agree on something on a call/etc :) then, you don't even have to think about serving context to AI
This sounds promising! Is there persistent memory along with the context layer? I've found that the memory aspect has been the biggest pain for me throughout my sessions, but the added context is definitely a game changer and something I've been looking for.
Definitely will try this out. Rooting for you guys!
Unabyss
@jaykhan3 thx Jay! And yes, there is persistent memory. For now, to save memory between sessions, you need to update the system prompt manually, e.g. in Claude.
We describe this in our guides, but it’s pretty straightforward and works well!
@marcin_uchacz1 amazing. if you can remove the need somehow to manually update e.g in Claude, that would be a gamechanger.
still love what you guys got cooking. would love to stay connected if you're on X + support where i can :)
@jaykhan on X
Connecting tools like Slack, Gmail, and GitHub to pull context is incredibly powerful but obviously raises immediate security questions. Can you talk a bit about how the data ingestion works? Is the raw text parsed and summarized entirely client-side, or is the indexing happening on your cloud servers?
Unabyss
@vikramp7470, everything is happening in our encrypted cloud. In transit from apps to our cloud and from cloud to you or your agent, your data is secure, too.