Blog

Development updates, tutorials, and tips

Multi Agent Plan Board
Develosaur

Multi Agent Plan Board

A multi agent plan board is one shared plan that several AI agents and the people around them read and write at the same time. What Develosaur's board actually does to make that work, the four mechanisms behind it, and one real session from digest to handoff.

Why I built Develosaur?
Develosaur

Why I built Develosaur?

I found myself in a quite strange position, where I felt way too confined, by some tools, and way too lazy for 80% of others, I'm talking about project management tools.

Turning a design doc into a board
Develosaur

Turning a design doc into a board

"Put this design doc into a develosaur board" turns headings into branches and everything else into their content, in one call, because a design doc is already organized the way a board wants it.

Persistent memory for AI agents vs a project board
Develosaur

Persistent memory for AI agents vs a project board

A memory server and a project board both answer "my agent forgets everything," but they store different things. What a memory layer remembers, what a board plans, and why an agent usually needs the second one more.

One local host, every agent, how the desktop host works
Develosaur

One local host, every agent, how the desktop host works

The desktop app is not a client. It is a small server you run once on your machine, and every local agent reaches your boards through it with nothing to configure per client and nothing to paste per repo.

ChatGPT and Develosaur, sandbox mode and connector mode
Develosaur

ChatGPT and Develosaur, sandbox mode and connector mode

Two different ways to connect ChatGPT to Develosaur, a no-account sandbox that builds a temporary board in one message and a signed-in connector for your real boards, and exactly what each one gets you.

Tags are not labels, they carry instructions
Develosaur

Tags are not labels, they carry instructions

A tag on a Develosaur board decides whether work is even visible to the ranked list, and can carry its own written rule that the server enforces or quotes back when it is broken.

I shipped Develosaur using Develosaur
Develosaur

I shipped Develosaur using Develosaur

The board that plans Develosaur's own engineering is a Develosaur board, worked by AI agents claiming tasks off it. What that forces you to get right, and the guardrails an AI-heavy codebase needs that a solo human team never would.

How the priority order decides what is next
Develosaur

How the priority order decides what is next

There is no scoring formula behind Develosaur's TODO list. Four layers in a fixed order, an objective dial, and a pinned head you control, and every ranked task says exactly which layer put it there.

Why Develosaur has no API keys to paste
Develosaur

Why Develosaur has no API keys to paste

The working design I said no to, why a bearer key you copy and paste was the wrong first feature even though it was the faster one, and what OAuth sign-in buys instead.

Several agents, one plan, how task claims stop them from colliding
Develosaur

Several agents, one plan, how task claims stop them from colliding

Point Claude Code, Codex and Cursor at the same Develosaur board and none of them touch the same task twice. How 45-minute task leases, the frontier and the abandoned-by trail work, and what happens when there is not enough ready work to go around.

Why a graph beats a list when an AI agent reads your project
Develosaur

Why a graph beats a list when an AI agent reads your project

An agent starting a fresh session knows nothing about your project. A list forces you to choose between dumping everything into its context or explaining it again. A graph lets it fetch what it needs, in order, and stop.

Keep the context!!!
Problems

Keep the context!!!

Whether by forgetting about specific .md, or getting too bloated, there's certain drift that happens if you don't track your agent or if you don't have a good source of truth, develosaur solves both memory, project tracking, and brainstorm board with one hit, a collaborative tool between humans and agents.

Capture tasks from anywhere
Develosaur

Capture tasks from anywhere

Most of the tasks, requests, or feature changes come through some feedback channels, or chats, emails, etc... it is very important to be able to on the spot catch these and place them so they don't get lost...