Skip to content

Free to start

Give Draftboard the whole list.

Paste one task or eighty. Director turns it into a plan you can check — duplicates merged, conflicts flagged, the right Character on each task — then runs it, and only comes back when it actually needs you.

Director is part of Draftboard, the workspace where your AI can actually work.

Illustration.

From your list to finished work

Four steps, and only one of them is yours. Auto is the default everywhere: you never need to know which Character, agent or effort a task needs — but you can always see it, and change it.

  1. 1

    Paste the whole list

    One task, a brain dump or eighty requests, in whatever order they come to you. Type it into “What needs done?”.

  2. 2

    Director organizes it

    Jev merges duplicates, splits big asks into tasks, orders what depends on what and flags requests that conflict.

  3. 3You

    You review the plan

    See every task with its project, Character, agent and effort. Change anything you like — or leave it on Auto, which is fine.

  4. 4

    Director runs it

    Independent tasks run at the same time, within your run budget. Results land in your workspace, and Director only asks when it needs you.

One shared Draftboard workspace

Plans, results, decisions and handoffs all land in the same place you already work.

The Characters

A team with names, not a single chatbot.

Each Character has one job, its own tools and its own permissions. Director decides who does what, and you always see who did it.

  • Forge

    Builder

    Can build the thing now.

    “Forge built it.”

  • Scout

    Researcher

    Goes and finds the answer.

    “Scout found it.”

  • Architect

    Planner

    Designs the plan before anything is built.

    “Architect planned it.”

  • Critic

    Reviewer

    Catches what's wrong before it ships.

    “Critic caught it.”

  • Scribe

    Writer

    Turns the chaos into clarity.

    “Scribe cleaned it up.”

  • Director Orchestrator

    Runs the team: plans the work and hands each part to the right Character.

  • Jev Chief of staff

    Turns a pile of requests into an ordered list, and asks when two conflict.

Multiple projects

One Director. Every project.

Different projects, different agents — and one place to tell them what needs doing. Codex on the web app, Claude Code on the API and the mobile app, Scout on research.

Director coordinates all of it: who works on what, what has to wait for what, and what can run at the same time.

Coding projects are folders on your computer, connected through Draftboard Companion.

Illustration.

Where the work runs

Some work keeps going without you. Coding waits for your computer.

Here's the difference, so you know what will be done when you come back.

  • On Draftboard

    Research, planning, reviews and writing

    Server work continues in the background, even with the tab closed. Changes to the canvas are applied when your workspace is open.

  • On your computer

    Coding with Claude Code or Codex

    Local coding runs through Draftboard Companion, only in project folders you've approved. If your computer is off or asleep, those tasks wait and start once it's connected again.

Stop being the middleman between your AI tools.

Before

  1. Claude says something
  2. ↓You copy it
  3. ↓Paste it into Claude Code
  4. ↓You wait
  5. ↓Copy the report
  6. ↓Send it to another model
  7. ↓Repeat ↺

With Director

  1. 1Jev and DirectorTake the whole list, not one prompt at a time.
  2. 2Route the workEach task goes to the right Character, agent and project — on Auto unless you change it.
  3. 3Artifacts and handoffsResults land in the workspace and are passed along, not pasted.
  4. 4Agents continueThe next step starts from what the last one produced.

One internal 80-request test run

What Director made of a messy list of 80

Duplicates, near-duplicates, vague asks, conflicting asks, later corrections, dependent work, research, reviews and coding — pasted in at once.

requests pasted in
80
duplicates merged
3
superseded by a later request
3
requests left to do
74
tasks from those requests
76
steps Jev added while working
6
tasks running at once, at most
4
tasks failed
0

An internal product test, not typical results: a fixed list of 80 requests run through Director's real planning and database, with a stand-in model and a simulated coding companion. It shows how Director handles a messy list; your own runs will differ.

What people hand to Director

  • Ship a feature

    1. Architect plans it
    2. Forge builds it
    3. Critic reviews it
    4. Tests verify it
  • Run a backlog

    1. Paste 20–80 requests
    2. Director removes duplicates
    3. Orders what depends on what
    4. Runs them
  • Run multiple projects

    1. Claude Code on the API
    2. Codex on the web app
    3. Claude Code on mobile
  • Research → build

    1. Scout researches
    2. Architect decides
    3. Forge implements

It changes the plan.

Director doesn't just execute the first plan. It updates the plan when the work changes.

When a step finds something missing, Jev adds the work it needs, puts it in the right order and picks the original step back up once it's unblocked.

You're still in control.

  • Nothing runs until you say so

    Director shows you the plan first. You can change any task, or run it as it is.

  • Approval where it matters

    Sensitive actions can be set to wait for your approval before they run.

  • Inside approved folders

    Coding agents work only inside the project folders you've approved on your computer.

  • Isolated worktrees

    Parallel coding in the same repo uses separate git worktrees, so agents don't edit over each other.

  • A budget for every run

    Pick Fast, Balanced or Best. Each run has limits on how many tasks it runs, how many at once and for how long.

  • A record of what ran

    Draftboard shows each step, who did it and why it was added.

Stop prompting one task at a time.

Give Director your backlog.

Free to start. Best on a desktop browser.