Your company’s AI brain,designed for every team,hosted in your GitHub.

Every team and every AI works from the same files: your decisions, research, playbooks and deliverables. Your non-technical teams read, edit and comment in a simple interface, while everything stays as markdown in your own GitHub repo, versioned and owned by you. Your technical teams keep a single source of truth they already know how to secure and back up.

Context is scattered and locked in one-to-one AI chats. And with a traditional vendor, your data, context and skills are locked in its proprietary databases, hard to retrieve and near-impossible to use anywhere else. With Lenti, everything is saved in your GitHub, in folders you can access at any time, including outside Lenti.

1. 1You connect a GitHub repo to Lenti. Your context lives there as markdown files and HTML artefacts.
2. 2Your teams browse, read, edit and comment in Lenti, from their browser, without a GitHub account.
3. 3Every change made in Lenti, by a person or by their AI, is a git commit in your repo, authored by that person.

Your technical teams connect GitHub once, and access for everyone else is managed in Lenti. Each team gets exactly the folders it needs, in read or write, and its members inherit those rights. You have a single token to monitor, renew or revoke, instead of one GitHub account and one set of permissions per person.

Giving a non-technical team access to GitHub means creating accounts, paying for seats and setting permissions person by person, a job that falls on your technical teams. GitHub sets rights per repo, so giving someone access to a folder means giving them the whole repo.

1. 1An admin connects a single fine-grained GitHub personal access token (PAT), limited to the repos you choose.
2. 2Admins create teams and give each team or member read or write rights, folder by folder. The most specific rule wins, so a read-only folder stays read-only inside a repo open for writing.
3. 3Lenti applies the same rights in the browser and over MCP, so a member's AI can only read and write what that member can.

All your skills sit in a single library in your repo, instead of being scattered across each person's chats and AI settings. Categories group them by team, function or use, and decide who sees what, so your sales skills reach sales and your finance skills reach finance. Your AI loads the right skill from the library when it needs it, so every team works from the current version.

Outsourcing your skills library to a single AI vendor is risky, and maintaining several native libraries across several AIs is inefficient. If you already host your skills in GitHub, seeing which ones are used most isn't obvious, and your managers have no way to manage by usage, the model that is proving itself for surfacing and promoting your best ways of working.

1. 1Skills are markdown files in your library. Standard skills belong to the organisation. Personal skills belong to their author, and every member can read them.
2. 2Admins sort standard skills into categories and give each team or member None or Read access per category. Only admins edit standard skills.
3. 3Over MCP, your AI lists and loads only the skills the person can read. In Lenti, filters by category, ownership and member make the library easy to browse.

Usage shows you which files and skills your teams and their AI actually rely on. What nobody uses is easy to spot and retire, so your AI stops working from stale material. The personal skills your colleagues adopt become standard for everyone, and the best ways of working spread across teams instead of staying with their author.

Context piles up, and your AI reads a file last updated a year ago with the same confidence as one written this morning. Good skills stay in one person's space while the shared library fills up with skills nobody uses.

1. 1Lenti counts real uses: a context file read by an AI over MCP or opened in Lenti, a skill loaded by an AI, an artefact opened. Each use counts once per person, per item and per day.
2. 2The Frequency view ranks items by uses over the day, week or month. The Recency view groups them by last use, down to never used. Filtering by last modified surfaces heavily used files that nobody has updated.
3. 3An admin promotes a proven personal skill or artefact to standard in a single action. If it stops earning its place, the admin moves it back to its author's personal space.

Your teams keep working in the AI they already use, whether Claude, ChatGPT, Grok or any other MCP client, with no new AI to adopt. Your context and skills live in your repo, not with an AI vendor, so changing model or tool later costs nothing: you connect the new one, and it reads the same memory.

Every AI vendor wants your context inside its own product, in projects, memories and custom instructions that don't move with you. The more you invest in one vendor, the harder it becomes to switch when a better model comes out.

1. 1You add Lenti as an MCP server in your AI client and sign in to Lenti once, in your browser.
2. 2For clients without that sign-in flow, such as scripts and automations, each member can create a personal bearer token in Lenti and revoke it at any time.
3. 3Either way, the AI acts on the person's behalf, with their rights, and every write it makes is a commit in their name.

Every activity runs on the skills that define how it's done, so the same activity produces the same kind of deliverable, whoever runs it and whichever AI they use. Progress rolls up from deliverables to activities to the project, so you see where a project stands without having to ask. A skill that reliably produces a deliverable in session is also one you can later hand to an agent.

Project management tools are built to be updated by hand, while more and more of the work happens in AI chats they never see. Deliverables vary from one person to the next, and reporting means chasing everyone for their status.

1. 1A project has a goal and activities. Each activity is a folder in your repo, bound to one or more skills that set the method and the definition of done.
2. 2Your team asks their AI to work on an activity. Over MCP, the AI reads the project, the activity and its skills, then writes the deliverables into the activity's folder.
3. 3Tracked deliverables and activities carry a To do, Doing or Done status, updated by a person or by their AI. An activity moves to Done when all its deliverables are, and a project when all its activities are.