> ## Documentation Index
> Fetch the complete documentation index at: https://docs.onelamp.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Shared memory for your AI. What each tool learns, every other tool can draw on.

OneLamp is the **shared context layer** for all your AI tools. You use the best AI
tool for each task (Claude Code, Codex, Cursor, ChatGPT, whatever ships next) — but
they don't share what they learn, so moving context between them falls to you. OneLamp
carries what each tool learns to the next over a single MCP endpoint, so every tool
works from what the others already know.

## The problem

You use the best AI tool for each task, but they don't share what they learn. What
Claude learns, Gemini never sees; ChatGPT's memory never reaches your editor's tool.
So moving context between them is manual — that's you, re-explaining yourself at
every handoff.

And the barrier is structural: an AI provider's memory is single-provider *by definition*,
and no one is incentivized to bridge the silos, so they stay permanent.

**Only a neutral layer can be cross-provider.**

## How OneLamp solves it

OneLamp sits across all your AI tools as one context layer. Every tool contributes
what it learns; every tool uses what the others already know. **Teach one, and
they all learn**, so you never have to repeat yourself.

<CardGroup cols={2}>
  <Card title="Connect once, use everywhere" icon="plug">
    Link your tool to one personal MCP endpoint. Every tool that speaks MCP reads
    and writes your context through the same tiny surface.
  </Card>

  <Card title="Retrieval, not generation" icon="magnifying-glass">
    The default query path returns a ranked pack of *your* material to reason
    over — never a synthesized answer that drifts from the source.
  </Card>

  <Card title="Cross-provider by design" icon="arrow-right-arrow-left">
    Move from Claude to Gemini to Codex and the next tool already knows you. Your
    context follows you, not the other way around.
  </Card>

  <Card title="You own your data" icon="lock">
    Your context lives in your own per-user store. Export the whole thing as
    portable JSON at any time.
  </Card>
</CardGroup>

## How it works

<Steps>
  <Step title="Sign in">
    Create a OneLamp account at [app.onelamp.ai](https://app.onelamp.ai) with
    GitHub, Google, or email.
  </Step>

  <Step title="Connect a client">
    Copy your personal MCP endpoint and add it to any AI client. The client
    signs in to OneLamp once via OAuth; no API keys to manage.
  </Step>

  <Step title="Seed it once (optional)">
    Run `/ol`, or point any tool at [app.onelamp.ai/setup.md](https://app.onelamp.ai/setup.md),
    to pull in what a tool *already* knows about you. See the
    [Quickstart](/quickstart#4-seed-your-context-one-time).
  </Step>

  <Step title="Work as usual">
    Each tool loads relevant context at the start of a task and saves durable
    learnings as it goes — automatically in clients that support hooks. The next
    tool uses all of it.
  </Step>
</Steps>

## What's live today

OneLamp is in **closed alpha (Phase 0)**. The shipping product is the
personal MCP endpoint and its tools:

* **Context memory** — `save_context`, `save_session`, `get_context`,
  `list_context`, `resume_session`, and `forget_context` give any tool durable,
  portable context. These six memory-CRUD tools are the entire MCP surface. See
  [Context memory](/concepts/context-memory).
* **Library** — ingest documents and URLs into an interlinked,
  tool-maintained Library you can query, managed from the OneLamp web app. See
  [Library](/concepts/library).
* **Data sources** — connect your other tools (Notion, Drive, Slack) over MCP and
  index their content on demand into your context, from the web app's Data Sources
  page. See [Data sources](/concepts/data-sources).

<Note>
  New to OneLamp? The [Quickstart](/quickstart) gets you from sign-up to a
  connected tool in a few minutes.
</Note>

## Principles

<CardGroup cols={3}>
  <Card title="Tool-agnostic">
    Models come and go. Your context shouldn't.
  </Card>

  <Card title="Zero friction">
    If it takes effort to maintain, it's already failed.
  </Card>

  <Card title="Privacy-first">
    Your context is yours. Period.
  </Card>
</CardGroup>