What your agents read before they render.
A playbook is what the team decided before anyone watched: who it’s for, what every version must do and never do, the frames that show what right looks like, and the steps for recurring work. Every agent reads it first, so notes only catch what nobody could decide in advance.
Agents readrevisions in force: House r3 · Northwind r6revisions in force: House r3 · Northwind r7
Saved — agents read it from now on
From the House down, the deeper one wins.
The House playbook is the studio’s, and every project and folder can add its own. A folder reads the House, then its project, then itself; where they disagree, the deeper one wins, and every part says where it comes from, so nothing is merged silently.
- Only people edit: agents, scripts and API tokens may only suggest, and MCP has no editing tool.
- Every save is a revision, with who, when and what changed; History shows the diff.
- Nothing is overwritten: if someone saved the same section while you wrote, you keep theirs or save yours over it.
- Only for the team: review links never reach a playbook, so clients don’t read it.
Show what right looks like. Write down how it’s done.
References are pictures, links and moments of approved versions, each frame taken frame-exact. Skills are the steps for recurring work in the open Agent Skills format, with their presets, LUTs or scripts: agents download them, and nothing runs in Lampo.
caption heightEvery mile, on the record.northwind-social.mp4 · V1 · 00:06:06- The end cardV2 · 00:12:12
- Caption height on ReelsV1 · 00:06:06
- Northwind colourspicture
- Brand guidelineslink
launch-editReading the skill export-reels
- A SKILL.md: a name, when to use it, then the steps. Import one you have, or write it in the app.
- Files are never run: they are downloads, used by the agent on its own machine under its own rules.
- A picture is a still: anything with motion is refused, and a link loses any credentials in it.
- Room for a studio: 30 skills per playbook, 10 files of up to 2 MB per skill, 24 references.
Read before every render, one line when nothing changed.
Agents get the playbook merged, deepest first, with the revisions in force, and every note they read names those revisions, so they see when the rules change. Asked again with what they read, Lampo answers in one line. Each new version is stamped with the revisions it was made under.
Agents suggest. A person decides.
When the notes keep asking for the same thing, an agent can suggest a rule, with its reason and the notes behind it. It waits in the inbox and in the section it changes, as a diff. Accepted, it is the next revision; rejected, the agent reads why.
- 00:13:20ShouldMiaThe logo comes too early: let it land with the claim.northwind-launch.mp4
- 00:14:02ShouldJo · clientLove it. Can the logo land a beat later?northwind-launch.mp4
- 00:05:10ShouldMiaLogo after the claim here too.northwind-social.mp4
Accepted — the playbook says it now
What the notes taught, on one page.
Nobody writes it. Lampo sums up a project’s notes in one short page an agent reads before it renders: what you keep asking for, what you love, the decisions that stand and the fixes that worked. The same notes always give the same page.
Try Lampo free.
Lampo Cloud from €0, or the open‑source app on your own machine or server. Version 0.1. The source is on GitHub.