Hey everyone,
We’ve been working on a project called Tizbe.
The simple idea: Tizbe is a shared memory layer for AI work. It lets you save project context and continue it across different AI tools like ChatGPT, Claude, Perplexity, Gemini, different accounts, and collaborators, without constantly copy-pasting or starting from zero.
There are two ways to use it:
MCP mode: you connect Tizbe to an AI client that supports MCP. Then the AI can read from and save to a chosen memory space during the chat. For example, you can start research in ChatGPT, continue in Perplexity, then refine in Claude, while all useful context is saved into the same place.
Manual mode: you can also save or update memory directly through the Tizbe web app. This is useful when a provider does not support MCP yet, or when you want to curate the context yourself.
This also makes collaboration possible. Different people can work on the same project memory, explore different directions in parallel, compare what each side found, and then merge useful branches back into one shared understanding or document.
We decided to build it on ICP as an experiment. Partly because the data model fit what we needed: persistent memory spaces, saved points, public/private access, and a history of how knowledge evolves over time.
To be transparent, we are still discovering the cost side. It is possible we eventually move some parts to a more traditional web2 stack if ICP costs are not sustainable long term, but for the current version ICP has been a useful place to test the model.
We’ve been using Tizbe internally for a while. Now a few local companies are starting to use it too. One of them begins onboarding workers in the next few days, and we expect around 1,000 users from these companies. They are paying us off-chain for now.
That’s it. Just wanted to share what we’re building and why we tried ICP for it.
Happy to hear feedback.
Peace.