# OneLamp — one-time setup prompt

Paste this into any AI agent that's connected to your OneLamp MCP server, or tell
the agent: **"Fetch https://app.onelamp.ai/setup.md and follow it."**

It does two things, once: (1) seeds your OneLamp context from everything
this agent already knows about you, and (2) adds OneLamp to this agent's system
prompt so it keeps loading and saving context going forward — important for agents
that *don't* have lifecycle hooks (browser Gemini, Perplexity, ChatGPT, etc.).

---

## Part 1 — Seed your context

1. **Recall what you already know about me.** Review my preferences, goals, the way
   I work, the decisions I've made, my projects, and my environment — from our past
   conversations, this project, and any memory you hold.

2. **Scan this workspace for existing agent-memory files** and fold them in if
   present: `CLAUDE.md`, `AGENTS.md`, `.cursorrules`,
   `.github/copilot-instructions.md`, `.windsurfrules`.

3. **Save each durable fact to OneLamp.** For every reusable fact, call the
   `save_context` tool — one fact per call — with:
   - `content`: the fact, in my words where possible.
   - `kind`: `fact` | `decision` | `preference` | `note`.
   - `topic`: a stable key (e.g. `deploy-schedule`, `package-manager`) so a newer
     fact supersedes an older one.
   - `source_surface`: the name of this agent (e.g. `claude-code`, `chatgpt`,
     `cursor`, `gemini`, `perplexity`).

   Skip secrets, credentials, and one-off chit-chat — only what's durable and
   reusable.

## Part 2 — Add OneLamp to your system prompt

So you keep using OneLamp on every future task (most important if you have **no
lifecycle hooks**), add this to your system prompt — the always-on instructions you
read on every task:

- **If you can edit project files** (Claude Code, Cursor, Copilot, Codex,
  Windsurf): append the block below to the instructions file this project uses —
  `AGENTS.md`, `CLAUDE.md`, `.cursorrules`, `.github/copilot-instructions.md`, or
  `.windsurfrules`. Don't duplicate it if it's already there.
- **If you're a browser agent and can't edit files** (Gemini, Perplexity, ChatGPT
  web): you can't set it yourself — show me the block and tell me to paste it into
  your **custom instructions / saved info / personalization** settings.

System-prompt block to add:

```
## OneLamp (connected via MCP)
At the START of every task, call the OneLamp `get_context` tool with a short query
about the task, to load what my other agents have already learned about me.
When you learn something durable (a decision, preference, fact, or convention),
call `save_context` (set `source_surface` to your own name). At the END of a
session, call `save_session` with a brief digest of everything durable you learned,
so nothing is lost. OneLamp is my shared context across every AI agent —
read from it and write to it.
```

## Finish

Tell me: how many items you saved (grouped by `kind`), and where you added the
system-prompt block (or paste it for me to add). From now on, every agent I
connect can call `get_context` and start already knowing this — no re-explaining.