Skip to content

Free desktop app

Delegate to AI. Keep track.

Delegate to AI. Keep track.

Project Orchestrator lets AI assistants work on your projects while you stay in control: plans, a memory of what was decided, and your approval when it matters. Today is the one screen that shows what needs your answer, what to pick up again, what is running and what comes next.

Version 0.0.19Also formacOSWindowsLinux

Free · MIT license · Bring your own AISee what it costs

  • MIT license, with Enterprise Edition modules under BUSL-1.1
  • Free to download. Needs Docker Desktop. Bring your own AI: Claude Code (the default) or another engine.
  • macOS, Windows, Linux

A request from an assistant arrives, you approve it, the assistant gets back to work, a task is completed and the plan moves forward.

Try it: pick an answer

Today · Sunday 4 October

An assistant is waiting for your answer

It has been waiting for 1 min.

45 %

17/38 tasks done

across 3 plans tracked here

  • 1 to pick back up
  • 3 to read

Waiting for you1

  • QuestionWelcome dinner and menuEvents

    The caterer offers two menus for 24 guests. Which one should I confirm?

    Free conversation: not attached to any plan

In progress2

  • Organise the autumn team offsite1h 16m$5.85
    10/183 in progress
    3 assistants are working on itEvents
  • Welcome dinner and menu28m$2.05
    2/81 waiting for your answer1 in progress
    1 task is waiting for your answerEvents

Assistants

1 waiting for your answer · 1 working

$5.50

Illustration of the app’s Today screen. The names are made up.

Download

Download Project Orchestrator

One-click install

1 click: download, open, connect your AI.

There is no Project Orchestrator account to create.

Free · MIT license · Bring your own AISee what it costs

Version 0.0.19 · 13 Nov 2026

The app installs on a computer. Pick the one you use.

Prefer a terminal?

Homebrew · Mac and Linux. Installs orchestrator, orch and mcp_server.

brew install this-rs/tap/project-orchestrator

What you need

  • Docker Desktop

    Runs the database and the search engine the app relies on. The app starts them for you. Get Docker Desktop

  • Claude Code

    The default engine for chat. The AI itself is not included: with Claude Code, you use your own Claude account, and the app can install it for you. Other engines (Codex, an OpenAI-compatible model, an ACP agent) are available from Settings — newer, in alpha. Get Claude Code

  • At first launch

    A setup wizard in four short parts checks Docker, installs Claude Code if it is missing and downloads the search model.

Install Docker and Claude Code from this page

Docker Desktop

Runs the two services that hold the project memory (Neo4j and Meilisearch).

Open the downloaded .dmg, drag Docker into Applications, then open Docker once and wait for the whale icon to stop moving.

Or in a terminal

brew install --cask docker

Check that it works

Each line should print a version number. If you see « command not found », open a new terminal window and try again.

docker --versiondocker compose version

Docker must be running (open it once) before you start Project Orchestrator. Installation page of the publisher

Claude Code

The assistant Project Orchestrator works with. It uses your own Claude account.

Paste the command in a terminal. It installs everything it needs and asks for your password only if required.

Command

curl -fsSL https://claude.ai/install.sh | bash

Check that it works

Each line should print a version number. If you see « command not found », open a new terminal window and try again.

claude --version

Run claude once to sign in. Installation page of the publisher

Free · MIT license · Bring your own AI See what it costs

Try it: start a task in any window

  • My day7

    • Confirm the venue deposit
    • Send the save-the-date to guests
    • Pick the menu for the welcome dinner
    • Book the train tickets
    • Print the name badges
    • Brief the photographer
    • Prepare the closing speech

    macOS

    Version 0.0.19 · 87.1 MB

    .dmg

    Download .dmg
  • My day7

    • Confirm the venue deposit
    • Send the save-the-date to guests
    • Pick the menu for the welcome dinner
    • Book the train tickets
    • Print the name badges
    • Brief the photographer
    • Prepare the closing speech

    Windows

    Version 0.0.19 · 88.7 MB

    .exe.msi

    Download .exe
  • My day7

    • Confirm the venue deposit
    • Send the save-the-date to guests
    • Pick the menu for the welcome dinner
    • Book the train tickets
    • Print the name badges
    • Brief the photographer
    • Prepare the closing speech

    Linux

    Version 0.0.19 · 163 MB

    .AppImage.deb.rpm

    Download .AppImage

Three steps, and you are in

  1. 1

    Download

    Pick the file for your system above. There is no account to create.

  2. 2

    Open

    Open the file. A setup wizard checks Docker, installs Claude Code if it is missing and downloads the search model.

  3. 3

    Connect your AI

    Chat runs on Claude Code by default; other engines are available from Settings (alpha). If it asks you to sign in, type this once in a terminal.

    claude
Desktop platforms
macOS, Windows, Linux
License
MIT, with Enterprise Edition modules under BUSL-1.1
Other systemsEvery file for Mac, Windows and Linux, with its size.

macOS

  • macOS, Apple Silicon

    The app for Macs with an M1 chip or later.

    Project.Orchestrator_0.0.19_aarch64.dmg · 87.1 MB

    .dmg
  • macOS, Intel

    The app for Macs with an Intel chip.

    Project.Orchestrator_0.0.19_x64.dmg · 89.8 MB

    .dmg
  • macOS, Intel (macOS 10.15 to 13.3)

    For Intel Macs on macOS 10.15 to 13.3. Without local embeddings: use an HTTP provider such as Ollama.

    Project.Orchestrator_0.0.19_x64-legacy.dmg · 47.7 MB

    .dmg

Windows

  • Windows, installer

    The usual installer for 64-bit Windows. It opens a setup wizard.

    Project.Orchestrator_0.0.19_x64-setup.exe · 88.7 MB

    .exe
  • Windows, MSI

    The same app as an MSI package, for managed deployments on 64-bit Windows.

    Project.Orchestrator_0.0.19_x64_en-US.msi · 120 MB

    .msi

Linux

  • Linux, AppImage

    One file that runs on most Linux distributions (x86_64).

    Project.Orchestrator_0.0.19_amd64.AppImage · 163 MB

    .AppImage
  • Debian and Ubuntu, x86_64

    Package to install with dpkg or apt.

    Project.Orchestrator_0.0.19_amd64.deb · 98.1 MB

    .deb
  • Debian and Ubuntu, arm64

    Package to install with dpkg or apt, for ARM64 machines.

    Project.Orchestrator_0.0.19_arm64.deb · 94.4 MB

    .deb
  • Fedora and RHEL, x86_64

    Package to install with dnf or rpm.

    project-orchestrator-0.0.19-1.x86_64.rpm · 61.5 MB

    .rpm
  • Fedora and RHEL, arm64

    Package to install with dnf or rpm, for ARM64 machines.

    project-orchestrator-0.0.19-1.aarch64.rpm · 56.7 MB

    .rpm
For developersHomebrew, install script, Docker image and the orchestrator command line.

Command line

Homebrew

Mac and Linux. Installs orchestrator, orch and mcp_server.

brew install this-rs/tap/project-orchestrator

Install script

Mac and Linux. Latest release into ~/.local/bin. Add --no-frontend for the light variant.

curl -fsSL https://raw.githubusercontent.com/this-rs/project-orchestrator/main/install.sh | sh

PowerShell

Windows. Installs into %LOCALAPPDATA%\orchestrator\bin.

irm https://raw.githubusercontent.com/this-rs/project-orchestrator/main/install.ps1 | iex

Docker

Image for amd64 and arm64, with the web interface.

docker pull ghcr.io/this-rs/project-orchestrator:latest# API only, no dashboard: a version tag with the -api suffix# docker pull ghcr.io/this-rs/project-orchestrator:0.0.19-api

Server programs

orchestrator, orch and mcp_server, with the web interface included. Lighter versions without it are on the release page.

  • macOS, Apple Silicon

    Programs without the app window: orchestrator, orch, mcp_server

    orchestrator-full-0.0.19-macos-arm64.tar.gz · 81.9 MB

    .tar.gz
  • macOS, Intel

    Programs without the app window: orchestrator, orch, mcp_server

    orchestrator-full-0.0.19-macos-x86_64.tar.gz · 82.8 MB

    .tar.gz
  • Linux, x86_64

    Programs without the app window: orchestrator, orch, mcp_server

    orchestrator-full-0.0.19-linux-x86_64.tar.gz · 91.1 MB

    .tar.gz
  • Linux, arm64

    Programs without the app window: orchestrator, orch, mcp_server

    orchestrator-full-0.0.19-linux-arm64.tar.gz · 85.6 MB

    .tar.gz
  • Windows, x86_64

    Programs without the app window: orchestrator.exe, orch.exe, mcp_server.exe

    orchestrator-full-0.0.19-windows-x86_64.zip · 92.8 MB

    .zip

The desktop app starts Neo4j and Meilisearch in Docker for you. With the command line, one compose command does it: follow the quickstart.

More for developers
From download to first launchNo terminal needed. You install two helpers first, then the app does the rest and shows each step as it goes.

The screens you will see

  1. Step 1: The start screen

    What you should see. A list of checks. On the very first launch the Docker lines say “Setup pending”.

    What to do. Wait for the setup to open: the start screen only checks what is installed.

  2. Step 2: Setup 1 of 4: Infrastructure

    What you should see. A green line says “Docker Desktop is running” under two cards, “Docker (recommended)” selected.

    What to do. Keep “Docker (recommended)”. The app looks for Docker Desktop. When it says “Docker Desktop is running”, press Next. If it says Docker is missing or not running, use the button next to the message, then wait: the app checks again by itself. The services use the local ports 7474, 7687, 7700 and 4222, and the app itself uses 6600.

  3. Step 3: Setup 2 of 4: Authentication

    What you should see. Three choices (No authentication, Password, OIDC / OAuth), the first one selected.

    What to do. On a computer only you use, keep “No authentication” and press Next. On a shared or public network, choose “Password” and set an email and a password: without sign-in, the app’s server accepts connections from your whole network.

  4. Step 4: Setup 3 of 4: Chat AI

    What you should see. Claude Code CLI shows a green check once installed (until then: “Claude Code CLI is required” and an install button), and the search model is marked ready.

    What to do. If “Claude Code CLI is required” is shown, press “Install Claude Code CLI”. Then let the app download the search model when it offers to; the size is shown on screen. Next unlocks when both are ready. If you have not signed in to Claude yet, see “If it does not work” below.

  5. Step 5: Setup 4 of 4: Launch

    What you should see. A green “Configuration saved!” message and a “Restart Application” button.

    What to do. Press “Generate Config & Save”. When “Configuration saved!” appears, press “Restart Application”. A switch on this screen controls whether the app checks for updates at start-up.

  6. Step 6: Wait for the first start

    What you should see. A checklist (Docker Desktop, Neo4j, Meilisearch, NATS, Backend server) turning green one line at a time.

    What to do. On this second start the app downloads the three images it runs on, then starts them. Later starts reuse the images and are quicker. When every line is green, the main window opens. If a line says “Taking longer than expected”, wait. If a “Continue anyway” button appears, see “If it does not work”.

Your system

macOS

macOS 11 or later on Apple Silicon (M1 and later). On Intel: macOS 13.4 or later, or macOS 10.15 to 13.3 with the older-macOS file, which has no local embeddings.

Install and start Docker Desktop

Download Docker Desktop from Docker’s page, open the file, drag Docker to Applications, then open it and wait until it says it is running.

Get Docker Desktop

Docker Desktop is open and reports that it is running. It is Docker’s own screen, so it is not reproduced here.

Open the app

Open the .dmg, drag Project Orchestrator to Applications, then open it from there. If macOS says it cannot check the app, we could not verify how the build is signed: allow it in System Settings, Privacy & Security, with “Open Anyway”.

If it does not work

The three problems you are most likely to meet, with what the app shows.

Docker is not running

What you see. The app says “Docker Desktop is not running” or “Docker Desktop is required”.

What to do.

  • Open Docker from your Applications folder (the app’s own Open button does the same). Wait until Docker says it is running.
  • Come back to Project Orchestrator: it checks again every few seconds, for about two minutes. After that, quit and reopen the app.
A port is already in use

What you see. Neo4j or Meilisearch stays “Unhealthy”, or the app says it failed to start services. The services need the local ports 7474, 7687, 7700 and 4222.

What to do.

  • Close the other program using that port: another database, search server, or an older copy of Project Orchestrator.
  • Only the app’s own port (6600 by default) can be changed on the Infrastructure screen of the setup, “API Server Port”.

Only if you are comfortable with a terminal, this shows which program uses the port:

lsof -i :7687
The chat cannot reach Claude

What you see. The setup says “Claude Code CLI is required”, or the chat says it has no Claude login.

What to do.

  • On the Chat AI screen of the setup, press “Install Claude Code CLI” (or “Retry Install”).
  • If the chat then asks for a login, open the Terminal app, type the command below, press Enter and sign in to Claude.

Type this and press Enter:

claude
Windows

Windows, 64-bit.

Install and start Docker Desktop

Download Docker Desktop from Docker’s page, run the installer, then start Docker Desktop from the Start menu and wait until it says it is running.

Get Docker Desktop

Docker Desktop is open and reports that it is running. It is Docker’s own screen, so it is not reproduced here.

Open the app

Run the installer and follow the wizard, then start Project Orchestrator from the Start menu. Windows may show “Windows protected your PC” because no signing certificate is set in the build configuration: choose “More info”, then “Run anyway” if you trust the download.

If it does not work

The three problems you are most likely to meet, with what the app shows.

Docker is not running

What you see. The app says “Docker Desktop is not running” or “Docker Desktop is required”.

What to do.

  • Start Docker Desktop from the Start menu (the app’s own Open button does the same). Wait until Docker says it is running.
  • Come back to Project Orchestrator: it checks again every few seconds, for about two minutes. After that, close and reopen the app.
A port is already in use

What you see. Neo4j or Meilisearch stays “Unhealthy”, or the app says it failed to start services. The services need the local ports 7474, 7687, 7700 and 4222.

What to do.

  • Close the other program using that port: another database, search server, or an older copy of Project Orchestrator.
  • Only the app’s own port (6600 by default) can be changed on the Infrastructure screen of the setup, “API Server Port”.

Only if you are comfortable with a terminal, this shows which program uses the port:

netstat -ano | findstr :7687
The chat cannot reach Claude

What you see. The setup says “Claude Code CLI is required”, or the chat says it has no Claude login.

What to do.

  • On the Chat AI screen of the setup, press “Install Claude Code CLI” (or “Retry Install”).
  • If the chat then asks for a login, open PowerShell, type the command below, press Enter and sign in to Claude.

Type this and press Enter:

claude
Linux

AppImage for Intel/AMD; .deb and .rpm packages for Intel/AMD and ARM64.

Install and start Docker Desktop

Follow Docker’s page for your distribution, then start Docker and wait until it says it is running. The app also recognises Docker installed without the Desktop application.

Get Docker Desktop

Docker is running. It is Docker’s own screen, so it is not reproduced here.

Open the app

For the AppImage: in your file manager, open the file’s properties and allow it to run as a program, then double-click it. For a .deb or .rpm: open it with your software installer, then start Project Orchestrator from your applications menu.

If it does not work

The three problems you are most likely to meet, with what the app shows.

Docker is not running

What you see. The app says “Docker Desktop is not running” or “Docker Desktop is required”.

What to do.

  • Start Docker Desktop, or your Docker service if you use Docker without the Desktop application. The app’s Open button only tries to start Docker Desktop.
  • Come back to Project Orchestrator: it checks again every few seconds, for about two minutes. After that, close and reopen the app.
A port is already in use

What you see. Neo4j or Meilisearch stays “Unhealthy”, or the app says it failed to start services. The services need the local ports 7474, 7687, 7700 and 4222.

What to do.

  • Close the other program using that port: another database, search server, or an older copy of Project Orchestrator.
  • Only the app’s own port (6600 by default) can be changed on the Infrastructure screen of the setup, “API Server Port”.

Only if you are comfortable with a terminal, this shows which program uses the port:

lsof -i :7687
The chat cannot reach Claude

What you see. The setup says “Claude Code CLI is required”, or the chat says it has no Claude login.

What to do.

  • On the Chat AI screen of the setup, press “Install Claude Code CLI” (or “Retry Install”).
  • If the chat then asks for a login, open a terminal, type the command below, press Enter and sign in to Claude.

Type this and press Enter:

claude

How much does it cost?

The app is free. You only pay for the AI you already use.

Project Orchestrator has no price, no account to create and no plan to buy. The assistant it talks to is yours: your own Claude Code, signed in to your own account, or a model you connect. The only cost is the one you may already have with your AI provider.

  • Project Orchestrator

    0

    The app, the server, the code

    • The desktop app, the server and the source code are free to download and to use.
    • We found no Project Orchestrator account, plan, licence key or usage limit in the source.
    • No analytics or tracking code was found. The app only asks GitHub for new versions when it starts.
  • Your AI provider

    Your plan

    What you already pay, if anything

    • Chat runs through Claude Code on your computer, signed in to your own Claude account. We sell nothing on top: your plan stays yours, and what it covers is set by Anthropic’s terms.
    • The app shows what each assistant and each plan cost, and says where the figure comes from. An unknown cost is never shown as zero.
    • You can also connect another model in Settings. That has only been tried against test servers, so we promise nothing. Its provider sets its own price.
  • Your computer

    Docker Desktop

    Runs the database and the search

    • Docker Desktop has to be installed. The app starts and stops the database and the search engine in it for you.
    • Docker has its own terms of use: check them for your own use.
    • Your plans, notes and decisions stay on your computer. Only the chat goes out, to the provider you chose.

Your knowledge stays yours, and no tokens go to waste

  • What you and your assistants produce (plans, notes, decisions) is structured in a graph on your computer, not in an online service. You stay sovereign over it.
  • Before each answer, the app hands the assistant the notes, decisions and files that matter, once: you do not re-explain the project, and the same block is not sent again and again.
  • When a long conversation is compacted, the essentials come back in a short block; and each run shows what it cost and where the figure comes from.

Good to know

  • This is how it works today. What your plan covers, and whether it may be used this way, is set by your provider’s own terms, and they can change. Check them.
  • We quote no price for any provider: look at theirs.
  • Licence: the core is MIT. A part marked Enterprise Edition would fall under BUSL-1.1, where production use needs a commercial licence. We found no file marked that way in the source we read.

Built on open tools you can inspect

RustTokioNeo4jMeilisearchtree-sitterMCPAxumTauriReactNATS
  • 31

    MCP tools

    Each tool takes an action parameter: plan, task, note, decision, code…

  • 17

    tree-sitter grammars

    Rust, TypeScript, Python, Go, Java, C, C++ and more, parsed into the graph.

  • 3

    desktop platforms

    The Tauri app ships for macOS, Windows and Linux.

Live

What is moving right now

LiveUpdated now

Try: switch repository.

The problem

Three things go wrong when your AI assistants share no memory

Scroll through each one. First you see what happens. Then you see what Project Orchestrator keeps so it stops happening. The numbers on each screen are explained just below it.

  1. 1 / 3

    Your assistants start from zero in every conversation

    You explain your goals, your preferences and what was already decided. Then the chat closes and the assistant forgets it all. Next time, you explain it again.

    With Project Orchestrator

    PO keeps your notes and decisions with the project. Before a task, the assistant looks up what applies. It searches by meaning, not only by exact words.

    Under the hood

    Notes and decisions are stored in a graph in Neo4j. note(action: "search_semantic") finds them by meaning, with Meilisearch.

    Learn more

    What you are looking at

    1. What the assistant looks up before it starts: this task, linked to a note, a decision and two files.
    2. A note you wrote once. The assistant finds it by meaning, so you do not repeat it.
    3. A decision, with its date and the 3 tasks it affects.

    Try: click a node of the graph, hover a number in the list below, or switch between before and after.

  2. 2 / 3

    Assistants change the same thing at the same time

    Two assistants edit the same document. One starts a job before the job it needs is done. So you do one thing at a time, and you lose the time you wanted to save.

    With Project Orchestrator

    A plan says which task depends on which. Tasks that do not depend on each other go at the same time. The others wait for the tasks they need.

    Under the hood

    PO computes waves from the task dependencies with plan(action: "get_waves"). Tasks that declare the same affected files are split into different waves.

    Learn more

    What you are looking at

    1. Wave 1: tasks that use different files, so they run together.
    2. Wave 2 waits for wave 1 to finish.
    3. The plan states the order, so you know what runs now and what waits.

    Try: switch between before and after to see the two assistants collide, then the plan that keeps them apart.

  3. 3 / 3

    The reason behind your choices gets lost

    Why this place, why that budget. The reason stays in someone’s head or in an old message. The next assistant asks again, or undoes the choice.

    With Project Orchestrator

    A decision is saved with its reason and what it affects. The next assistant finds it, with its status, before it changes anything related.

    Under the hood

    A decision is linked to the tasks, plans or files it affects with decision(action: "add_affects"), and it has a status such as proposed or accepted.

    Learn more

    What you are looking at

    1. The assistant can look the proposal up against what was decided. An accepted decision says June.
    2. The decision, with its reason and its date. The assistant reads it before it changes anything related.

    Try: switch between before and after, or hover a number in the list below.

One plan, two views

Do several tasks at once, not one after another

Some tasks need another task to be done first. The others can start together, as a group called a wave. Watch the same plan as a map of links or as a board with columns.

Organize the team weekend

CompletedP98 tasks in 4 waves

8/8 done (100%)

  1. Wave 1 · 2 in parallel · 6 min

  2. Wave 2 · 2 in parallel · 14 min

  3. Wave 3 · 3 in parallel · 12 min

  4. Wave 4 · 1 in parallel · 7 min

Select a task to see what it waits for and what it unlocks.

The same plan on the same clock

One assistant, task by task, against the waves above.

Two tracks on one time axis. One assistant doing the tasks one after another needs 71 minutes. In waves, the independent tasks run together and the plan takes 39 minutes.

One assistant, task by task

71 min

In waves

39 min

Time saved

32 min

71 min one by one, against 39 min in waves: 45% less.

Try: drag a task to another column, click a node to light up its links, or slide the waves cursor.

Illustrative scenario: the durations are assistant minutes written for this demo, not a benchmark.

What it does

Your projects, plans and decisions in one place

Projects with or without code, plans, assistants that work together, a memory of what was decided, and sharing only with your consent. Each one is shown below.

  • Projects, with or without code

    A project can be a codebase, or only plans, notes and documents.

    Projects
    2 plans5 documentsno code
    1 plan3 documentsno code
    3 planscodebase

    Try: filter by type, then open a project.

    3 projects shown
  • Objectives, plans and tasks

    A goal broken into tasks, with what depends on what.

    Plans

    Organize the team weekend

    In progress
    1/4

    Try: click a task to move it forward and watch the progress.

    1 of 4 tasks completed
  • Assistants that work together

    Independent tasks go to several assistants at the same time.

    Run · 3 waves
    Pending
    Wave 13
    Venues
    Dates
    Budget
    Wave 23
    Booking
    Meals
    Invite
    Wave 32
    Schedule
    Share

    Try: press Run to play the waves one after the other.

    Wave 2 running
  • Notes and decisions

    What was learned and decided, kept with the project.

    Notes
    The weekend is 14 and 15 June
    Accepted
    booking, invitations
    Two people need vegetarian meals
    Active
    diets.xlsx
    The budget is 150 per person
    Needs review
    budget.xlsx
    Choose the mountain lodge
    Proposed
    booking
    Last year’s hotel is full in June
    Stale
    last-year.pdf

    Try: search the notes, then accept or reject the proposed decision.

    5 of 5 notes shown. Mountain lodge: proposed
  • Documents

    Attach files to your work and keep the originals with you.

    Documents
    budget.xlsx
    Spreadsheetteam-weekend
    venues.pdf
    PDFteam-weekend
    kickoff.pptx
    Presentationteam-weekend
    schedule-draft.docx
    Documentteam-weekend

    Try: filter by type or change the order.

    4 documents shown, sorted by date
  • Specialist assistants

    Assistants focused on one area, and knowledge that wakes up when needed.

    Skills
    event-planning
    Active
    EnergyHighCohesionHigh
    travel-booking
    Emerging
    EnergyMediumCohesionMedium
    old-vendor-list
    Dormant
    EnergyLow

    Try: use a skill, or let a week pass.

    event-planning active, travel-booking emerging, old-vendor-list dormant
  • Repeatable procedures

    A procedure the assistant follows step by step, every time.

    Protocol · publish-an-announcement
    Start
    Write
    Review
    Done
    RunningReviewgatePass

    Try: advance the run, or make the gate fail.

    State Review, gate Pass
  • Between PO instances, nothing is shared unless you approve it. Early version.

    Sharing & privacy
    Venue shortlist
    Waiting for your approval
    team-weekendManual
    The weekend is 14 and 15 June
    Approved
    expires in 7 days
    Budget notes
    Never shared
    budget.xlsxnothing was copied

    Try: approve, decline or revoke a share.

    Venue shortlist: Waiting for your approval. The weekend is 14 and 15 June: Approved. Budget notes: Never shared
  • Your choice of AI, or let PO choose

    Claude Code, another model, or Auto. You decide how much PO chooses.

    Providers
    Default
    Built in
    Allowed for team-weekend
    deepseek-chat
    Not allowed for this project
    SSH

    Claude Code: I split the weekend into four tasks. Nothing is booked yet.

    Try: choose the default assistant, allow or block a provider.

    Default: Claude Code. I split the weekend into four tasks. Nothing is booked yet.
  • Reachable from anywhere, on your phone too

    Put PO on your own domain, sign in with your account, and follow your projects from your phone.

    po.your-domain.com
    team-weekendAssistant online

    Plan the team weekend and find a venue.

    Three venues fit 12 people. The lake house is the cheapest. Shall I book it?

    Try: switch the device, sign out, answer from your phone.

    Signed in, seen from Phone

Plugs into anything

It plugs into your tools and remembers what works

Outside tools, procedures you write, and a memory that adapts. See what each one does today, and what it does not.

  • Your tools

    Connect MCP servers. Assistants call them under a policy that blocks calls that change things by default.

    Available

  • Your procedures

    Write a release, a review or a triage as states and triggers. Every run is recorded.

    Available

  • A memory that adapts

    Notes used together get linked and strengthened. Unused ones fade, and clusters become skills.

    Available

Try: open a card to see how it works.

What it does not do

Ask what it cannot do

Most tools list what they can do. You decide better when you know the edges first. Here is what Project Orchestrator does not do today.

  • It does not include the AI

    PO organizes the work and can choose between the models you connect. It does not supply the model: bring your own Claude account for Claude Code, or a key for another provider.

  • It needs Docker Desktop

    The database and the search engine run in Docker. The app starts them for you, but Docker has to be installed.

  • Other engines: available, in alpha

    Beyond Claude Code, run an agent on the native engine (any OpenAI-compatible model such as DeepSeek, or an endpoint speaking the Anthropic Messages API), on Codex, or on an ACP agent such as opencode. Codex is now reached for real — its own health check, the machine's login — and an ACP agent gets a system prompt, but a full turn over Codex or opencode has not been measured yet: that is why they are alpha. Claude Code stays the default.

  • Sharing between instances is an early version

    The approval step, the expiry and the revocation exist. The exchange between PO instances is still being finished.

  • It does not decide for you

    A run can wait for your review before it goes on. The decision stays yours.

  • It is not a phone app

    The app installs on a computer. On a phone, you reach PO through a server of your own.

  • A project with no code starts from an assistant

    The project form in the app does not offer that choice yet: you ask an assistant to create it.

Missing a limit you care about, or found one we got wrong?

Tell us on GitHub

Memory

PO remembers what you decide and learn

Notes, decisions and the links between them stay attached to your project. PO finds them again when they are useful. Scroll to watch a memory grow.

  1. 1

    Your files become a memory

    PO starts from what is in the project: documents, spreadsheets, files. Each one becomes a node you can find again.

  2. 2

    Plans and tasks join in

    Plans and tasks are added to the same picture. Each task is linked to the files it touches.

  3. 3

    Notes and decisions stay linked

    What you learn becomes a note. What you decide becomes a decision. Both stay linked to the files and to each other.

  4. 4

    Items about the same subject move close to each other. You see the main themes of the project at a glance.

  5. 5

    You ask a question. PO finds the notes that match (cyan). Then it follows the links to related items (violet), even when the words are different.

Memory

Invented example: a small event project. The names in the graph are illustrations, not your data. Colors and shapes are those of the product graph.

Under the hood: the names of the links

These are the names the product uses for the links in its graph. You do not need them to use PO.

  • CONTAINSA plan contains its tasks.
  • IMPORTSA file uses another file.
  • AFFECTSA task touches a file.
  • CO_CHANGEDTwo files that often change together, seen in the history of the code.
  • SYNAPSEA link between notes, decisions and files. Links nobody uses fade away.

How it works

From one sentence to checked work

Plan, order, recall, run, verify, learn. This is what PO does for every plan.

1 · Plan

You say what you want, you get a plan

You describe the result in one sentence. The assistant turns it into a plan with tasks. Each task has steps, and each step says how to check it is done.

2 · Order

Tasks that do not depend on each other go together

PO sorts the tasks. A task waits only for the tasks it needs. All the others can start now, at the same time.

3 · Recall

The assistant looks up what the project already knows

Before each task, the assistant reads the notes and decisions that apply. So it does not ask the same questions again.

4 · Run

Assistants do the tasks, group by group

Each task goes to its own assistant. The tasks of a group run at the same time, and the next group waits.

5 · Verify

A group is done when its checks pass

Before the next group starts, PO checks the work. The steps must be closed and no sensitive file may be left in the changes. For a code project, PO also checks that it still builds.

6 · Learn

What was learned is saved for next time

When the work is done, what was decided and learned is written back to the project memory. The next plan starts with it.

What stage 6 saves is what stage 3 recalls on the next plan.

Follow one run, stage by stage

Try it

Watch an assistant work with PO

Pick a request. You see the answer in plain words, and the calls the assistant makes to PO behind the scenes. Open a call to see its details.

Chat · team-weekendSimulated demo
Organize a weekend in June for my team of 12. We want a quiet place near a train station.

Two notes apply: the place must be close to a train station, and last year the menu had nothing vegetarian. I will write the plan around both.

The plan has 3 tasks in 2 waves. The shortlist and the date poll do not depend on each other, so they happen at the same time. Booking waits for both.

Scripted replies. No server is contacted.

Try: type your own request, open a call to see its details, then copy the whole exchange.

Simulated demo: the answers are scripted and no server is contacted. The calls and their arguments are the real ones.

Where PO is going

Each person keeps their PO. The organization shares what it chooses.

Think of a company: a common knowledge, teams, people and access rights. Part of the base runs today. The organization layer is planned.

Shared knowledgeOrganization

Illustration with invented people.

  • Available

    One PO per person: own identity, sign-in, workspaces, vault

  • In progress

    Sharing between PO instances, only with consent

  • Planned

    Common knowledge, teams and roles, access rights

Read the vision

Questions

Before you install

Short answers, each one checked against the source code of the app.

Is it really free?

Yes. The app, the server and the source code have no price, and we found no Project Orchestrator account, plan, licence key or usage limit in the source.

What you pay for is the AI, and only if your provider charges for it. The chat runs through the engine you choose: Claude Code and your own Claude account by default, or another engine. So the subscription or key you already have is the one in use. The app shows what each assistant and each plan cost. You also need Docker Desktop, which has its own terms.

Two reserves. What your plan covers, and whether it may be used this way, is decided by your provider’s terms, which can change: check them. And a part of the code marked Enterprise Edition would be under BUSL-1.1 (see “Which licence is it under?”); we found no file marked that way.

Where does my data go?

Your plans, tasks, notes and decisions stay on your computer, in a database and a search engine that the app starts for you. They only accept connections from your own machine.

Chat is the part that talks to the outside: what you type, and what the assistant reads to answer you, is sent to Claude by Claude Code, under the terms of Anthropic. If you connect another provider and allow it for a project, that project’s chat goes to that provider instead. The app keeps a journal of what was sent where, without the content.

Search by meaning runs on your computer (its model is downloaded once). The app checks GitHub for new versions when it starts. No analytics or tracking code was found in the source.

Do I need an account?

No account with Project Orchestrator: there is nothing to sign up for, and the setup uses no sign-in by default.

You can add a password, or sign-in with Google, Microsoft or another OpenID provider, during setup.

Good to know: with no sign-in, the app’s own server accepts connections from your whole network, not only from your computer. On a shared network, choose a password.

Do I have to install Docker?

Yes, by default. The app keeps your data in a database and a search engine, and it runs them with Docker Desktop. You install Docker Desktop once; the app then starts and stops those services itself.

The setup checks for Docker, tells you if it is missing or not started, and does not let you continue until it runs.

People who already run their own database and search servers can point the app at them instead (the “External servers” choice in the setup).

Do I need Claude Code and a Claude account?

Not necessarily. The AI is not part of Project Orchestrator: the app keeps your plans, notes and decisions, and a chat engine answers. Claude Code, Anthropic’s tool, is the default and the most tried path; it has to be on your computer and signed in to your Claude account, and the setup can install it for you.

That is also why there is nothing more to buy to orchestrate: the plan you already use is the one in use. Which plan Claude Code requires, and what it allows, is decided by Anthropic and can change: check their terms.

To sign in to Claude Code you may have to type claude once in a terminal; the app’s own message says so. No key is asked for in the setup. If you would rather not install it, choose “Another provider (configure later)” and add an engine in Settings, under Providers. Any engine other than Claude Code needs sign-in to be turned on.

Can I use it without writing code?

For daily use, yes: you work in the app (Today, plans, tasks, notes, decisions) and ask the assistant in a chat. The installation does not need a terminal either, apart from the possible one-time Claude sign-in.

A project does not have to be software: recent versions support projects without a code folder.

The honest limit: the product started with developers, and some screens still lean that way, for example the starter suggestions of the chat (impact analysis, code search, architecture).

Which computers does it run on?

macOS 11 or later on Apple Silicon (M1 and later); on Intel, macOS 13.4 or later, or macOS 10.15 to 13.3 with the older-macOS file (no local embeddings there), Windows 64-bit, and Linux (an AppImage for Intel/AMD machines, .deb and .rpm packages for Intel/AMD and ARM64).

Docker Desktop has to run on the same computer. There is no phone app and no hosted web version: the downloads on the release page are all desktop packages.

How do I update it?

The desktop app checks for a new version each time it starts and shows a banner when one is ready. You can switch this off in the last step of the setup.

If no banner shows up, download the latest installer from the release page and run it.

Can I use it with my team?

Not as a shared team space yet: today each person runs their own Project Orchestrator.

What exists: workspaces, to group several of your projects, and an early version of sharing between two instances. You decide, per project, whether notes are shared manually, suggested to you, or sent automatically; copies can expire, and you can withdraw a share.

What is planned and not available: one common knowledge base for a company, teams and hierarchy, roles and access rights.

Which licence is it under?

Two licences, as the LICENSE file states (copyright FFS SAS): MIT AND BUSL-1.1.

The parts under the MIT licence can be used, copied, modified and shared freely.

The part marked Enterprise Edition is under the Business Source License 1.1: you may test, develop and evaluate with it, but using it in production needs a commercial licence from FFS SAS (contact@ffs.holdings). That part becomes MIT four years after publication.

In the source we read, no file carries that Enterprise Edition mark.

What happens when I close the window?

Closing the window hides the app in the system tray; it keeps running. Choose Quit from the tray icon to stop it.

Quitting also stops the three services the app started in Docker (database, search, events). If you use your own servers, it leaves them alone.

How do I get help or contribute?

Join the Discord to ask for help or give feedback. For a bug, open an issue on GitHub and include your system, the steps to reproduce it and any error message.

Contributions are welcome. The contributing guide explains how to set up the project and what a good report contains; it is written for developers.

Which engines and models can I use?

Claude Code is the default and the most tried path. You can also run an agent on the native engine with any OpenAI-compatible model such as DeepSeek, on Codex, or on an ACP agent such as opencode. The conversation, tools and permissions are the same whichever engine runs it. You register an engine in Settings, under Providers, allow it project by project, and bring its key.

The honest limit: the engines other than Claude Code are alpha. They have been tried against stand-ins, never against the real services yet, so we do not promise they work. An engine other than Claude Code is refused unless sign-in is turned on.

Its price is set by its provider, and we quote none. The app shows each cost with where it comes from (reported, estimated, subscription, local or unknown) and never invents a zero.

What does Auto do, and can it override my choice?

On Auto, PO picks the model for each task, each delegation or, in full mode, each turn of a conversation, within hard limits on tools, context size, budget and consent. Every decision is recorded with a readable reason, and PO learns from the results.

A model you chose yourself is never replaced, even when a conversation resumes. In a conversation you can also tick the models you allow instead of using Auto.

The honest limit: routing has only run on two stand-in providers, never on a real model yet, and a pool of models that spans several providers is planned, not available. If a learned choice does worse than the declared one, routing falls back by itself to only recording what it would have chosen.

What happens if the server restarts in the middle of a task?

On the native engine, Codex and ACP, a turn cut by a restart resumes by itself, in the background, with a note telling the agent that its last tool call may or may not have run; a tool is never replayed. If the vault (where keys are kept) is locked, the resume waits for it. A turn that cannot resume is closed with a visible error instead of staying in progress. Claude Code, the default engine, is the exception: a turn cut by a restart is closed with a visible error, and your next message picks the conversation up (a Claude Code conversation whose process died takes that message instead of dropping it).

A conversation on the native engine keeps its transcript on disk and resumes after a restart too. The runs of a plan are saved, so a run can be resumed after a restart.

Can I approve a tool once, for the session, or always?

Yes. When an agent asks permission to use a tool, you can answer once, for the whole session or, in a project conversation, always. Before you click, the button shows exactly the call it covers.

Lasting rules are kept and checked by PO itself, on Claude Code and on the native engine, with a strict comparison (the identical call). They can be listed and revoked.

Which languages is the interface in?

The interface speaks 13 languages: English, French, Spanish, Portuguese, German, Chinese, Japanese, Korean, Vietnamese, Russian, Ukrainian, Hindi and Arabic. Translated so far: projects, plans and tasks, the intelligence screens, administration and the chat components, along with the shared layer.

The remaining screens follow, so some may still show in English.

Contribute

Got an idea, or hit a bug? Tell us.

Project Orchestrator is open source. Ideas and bug reports are both opened as issues on GitHub, where anyone can read them.

Request a feature

Describe the problem you want solved, and what you do today to work around it. A clear problem is easier to discuss than a finished solution.

It opens as a public issue on GitHub, where the maintainers and other users can read it and add to it.

Report a bug

Tell us what you did, what you expected and what happened instead. The more exact the steps, the easier it is to reproduce the problem.

It opens as a public issue on GitHub, where the maintainers and other users can read it and add to it.

A good report includes

  • The version of Project Orchestrator you run, for example the tag of a release on the Releases page.
  • Your operating system and its version.
  • The steps that lead to the problem, in order, and what you expected instead.
  • The exact error message or a screenshot, if there is one.

Get started

Give your agents a memory and a plan

Install the orchestrator, connect your assistant over MCP, and let your first agent read the project before it writes a line.

The app is free. You only pay for the AI you already use, such as your own Claude Code subscription or a key for another model. See what it costs

Or start from the terminal

brew install this-rs/tap/project-orchestrator