What to put in an agent’s memory on day one
The first thing most people do with an agent memory is load it: the project brief, the stack, the conventions, the people. A week later none of it has been read, because the agent had all of that in its prompt anyway.
Nothing. Assign it an empty bullet and give it one rule about what belongs there, because a memory you pre-fill is a memory it never learns to keep.
Set it up
- Point the client at https://pando.ink/mcp, which speaks streamable HTTP.
- Authenticate with OAuth 2.1 and dynamic client registration, or paste a key from Agents and API keys as a Bearer token.
The rule that is worth writing down
Keep decisions and their reasons, not transcripts. A decision is worth keeping when somebody would otherwise re-derive it: why the queue is durable, why that library was rejected, what the customer actually asked for. A transcript is worth keeping never.
What a good first week looks like
- Four to ten bullets, not forty. If it is longer than that after a week, it is keeping transcripts.
- Each one readable on its own. A note that only makes sense next to the conversation it came from is a transcript with a haircut.
- Reasons, not conclusions. "We use OPML because their importer chokes on Markdown" survives; "use OPML" does not.
- A recall before it assumes. The habit matters more than the content, and it only forms if the memory is worth reading.
Where to put the bullet
Inside the project it serves, not in a folder called Agent. A memory filed by owner is a memory nobody opens; a memory sitting under the project is one you read while you are already there. You can move it later without the agent noticing, because it follows the bullet rather than the path.
Deeper
Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this