The Seed Packet: Feeding Sera Her First Layer of Memory

The Seed Packet: Feeding Sera Her First Layer of Memory

A new hub knows nothing. That is the hardest moment in the whole product, because an assistant with no memory cannot demonstrate why memory matters. The Seed Packet exists to get you past it in one sitting.

Open /seed from the dashboard. It is admin-only.

---

How it works

Seven short steps, each one a small file about your organization. Every step card carries four things:

• An honest time estimate. Usually about ten minutes. Not "a few moments."
• Why Sera needs it. One paragraph on what this file lets her answer later.
• A copy-button prompt. Written to be pasted into whichever AI you already use, ChatGPT, Claude, Gemini, or Perplexity. It tells that AI exactly what to produce and what to leave out.
• A paste-back box. You paste the file it produced, and Sera extracts it into records.

Three ways to leave any step

Nothing traps you mid-step:

• Process now: extract this file into memory and move on.
• Save draft: keep what you have and come back to it.
• Skip: move past it entirely. You can return later.

Progress and drafts are stored per organization on the server, so you can start on a laptop, get interrupted, and finish from a different device. Nothing lives in one browser's memory.

The minimum viable seed is three steps

You do not have to do all seven. The first three are the core:

1. Organization Overview: who you are, where, for whom, and the few things that matter most right now.
2. Current Projects: the three to five active ones, what done looks like, who leads each.
3. Key People and Organizations: who is involved and how they relate to the work.

Finish those and Sera can already answer real questions. The remaining steps, decisions, tasks and commitments, risks and tensions, and recent meetings, deepen her rather than unlock her.

Your AI does the writing

The prompts are designed so the heavy generation happens in your own AI, on your own account. That is deliberate: it keeps the cost of seeding predictable, and it means you review what your AI produced before any of it reaches Sera. Saberra pays only for one clean extraction pass per step you choose to process.

The guidance on the page is static text, so reading the packet costs nothing at all.

The written guide is right there

The full written prep guide is embedded inline on the same page. There is no Drive link to find, no PDF to download, and no external document that could go stale in a different place. If you would rather read the whole thing and prepare files your own way, it is one click away, and the article "Data Seeding Prep Sheet" is the same material in the Knowledge Base.

---

The one rule has not changed

Do not seed everything you have. Seed what is current, reviewed, and useful.

Old archives, abandoned plans, and unreviewed dumps create noise that every future answer has to wade through. Three clean files beat thirty messy ones, and the packet is deliberately shaped to make the small, clean version the easy path.

Before you paste anything back, ask the three questions: is this current, has a human read it, and would it help Sera answer a real question later?

What you get at the end

A live counter on the page shows how many records Sera now holds, so the progress is visible rather than promised. The completion card offers three starter questions built from what you actually seeded, which is the fastest way to feel the difference between an empty hub and a useful one.

From there, memory maintains itself: meetings and email flow in on their own, and the seeding was only ever the first layer.

Key points

Open /seed in the dashboard (admin only). Seven short steps, each carrying an honest time estimate, a plain explanation of why Sera needs that file, a copy-button prompt written for whichever AI you already use, and a paste-back box. Each step has three exits: process it now, save a draft, or skip it. Progress and drafts persist per organization on the server, so you can leave from one device and resume on another. The first three steps are the minimum viable seed: Organization Overview, Current Projects, and Key People and Organizations. A live counter shows how many records Sera now holds, and the finish card offers three starter questions built from what you just seeded. Your own AI does the heavy generation, which keeps cost predictable, and Saberra pays only for one extraction pass per processed step. The full written guide is embedded on the page as a bypass, so there is no Drive link or external document to chase. The one rule still applies: seed what is current, reviewed, and useful, never everything you have.