Skip to content

gitmarcello/dduo-solo-founder

v0.2.0-beta.2Apache-2.0

Isolated project memory, planning, and task context for Codex and Claude Code

EN · English · IT · Italiano

dDuo Solo Founder

Give Codex and Claude Code project memory inspired by human memory.

dDuo retrieves relevant information through embeddings and consolidates it during “sleep”: it processes conversations to retain decisions, lessons and useful memories for future chats.

It also provides a project dashboard: while you work in chat, the assistant can create and update tasks, organize plans, epics and sprints, and keep the backlog and completed work tidy. In the same dashboard, you can explore the memory and see how much context and usage dDuo adds.

The system runs on your computer or your server, with separate memory for each project, whether you work alone or with a team. Semantic search uses embeddings from the OpenAI API.

Start here

Install dDuo for this project

Open the project you want to work on in Codex or Claude Code and paste:

Install this plugin in this project and guide me through setup:
https://github.com/gitMarcello/dduo-solo-founder/tree/v0.2.0-beta.2

The assistant follows this release's installation instructions and security guide; no long prompt is needed.

Prefer the terminal? Installation command

Run from your project's root. Requires Node.js 18+ and Git:

dduo_install_dir="$(mktemp -d)"
git clone --depth 1 --branch v0.2.0-beta.2 https://github.com/gitMarcello/dduo-solo-founder.git "$dduo_install_dir"
"$dduo_install_dir/install.sh" --only codex --project-root "$PWD" --yes
# Use --only claude instead when installing for Claude Code.

On native Windows, use PowerShell:

$dduoInstallDir = Join-Path ([IO.Path]::GetTempPath()) ("dduo-" + [guid]::NewGuid())
git clone --depth 1 --branch v0.2.0-beta.2 https://github.com/gitMarcello/dduo-solo-founder.git $dduoInstallDir
node (Join-Path $dduoInstallDir 'bin/install.mjs') --only codex --project-root (Get-Location).Path --yes

The assistant runs the installation and opens Setup. Complete the actions shown there, then follow the restart instructions. See requirements.

Create the first memory directly on a VPS

A new project's memory can live on the VPS from the beginning, without local Docker. The assistant follows the headless VPS runbook.

Set up this project's first dDuo memory directly on a VPS and connect this
computer remotely. Preserve any existing memory, check the server requirements
and follow the procedure here. Ask only for missing details and decisions:
https://github.com/gitMarcello/dduo-solo-founder/blob/v0.2.0-beta.2/docs/platform-support.md

Move an existing dDuo project to a VPS

Paste this in the existing project's chat. Other projects keep their own setup.

Move this project's existing dDuo memory to a shared VPS. Before handling
infrastructure credentials, enable and verify off-record capture. Verify a
recovery backup, keep credentials private and maintain one writable memory.
Follow this release's transfer and recovery procedures; ask only for missing
details and necessary decisions:
https://github.com/gitMarcello/dduo-solo-founder/blob/v0.2.0-beta.2/docs/remote-teams.md
https://github.com/gitMarcello/dduo-solo-founder/blob/v0.2.0-beta.2/docs/backup-and-recovery.md

Join a project someone already shared

Use the manager's one-time invitation in place of the local installation prompt in your authorized Git checkout. It connects to the existing memory without local Docker or VPS credentials. Memory access does not grant Git access.

What you get

  • Project continuity: retrieve relevant decisions, constraints and lessons across chats, with sources and revision history.
  • Organized work: discuss the goal; the assistant can propose a Plan, group work into Epics, plan Sprints and keep Backlog and completed history separate, with readable Task links.
  • Shared procedures: keep testing, deployment and working rules in a versioned operating manual, useful both alone and with a team.
  • A dashboard you can inspect: read memories, follow work, inspect delivered context and usage measurements, and manage encrypted recovery backups.

For example: “Pick up where we left off and suggest the next step”, “Turn this idea into a plan” or “What have we already decided about authentication?”

Requirements

ScenarioMinimum
Shared VPSLinux with systemd, 1 vCPU, 1 GiB RAM, 2 GiB persistent disk-backed swap, 5 GiB free in Docker storage after swap

On macOS or native Windows, install Node.js 18+, Git and your supported client; local memory also uses Docker Desktop. No local hardware minimum is imposed. The installer sets up uv. The person hosting the memory supplies an OpenAI API key for embeddings (semantic search) and one supported subscription login: Codex or Claude. The first supported project chat establishes the preferred sleep provider; before each consolidation the memory host verifies that preference and can use the other already-verified host subscription only if the preferred one is unavailable. Collaborators supply their own chat client access and use the host's memory services.

The VPS minimum is for one lightly loaded project. Every additional project keeps its own complete stack and therefore needs additional RAM and storage. remote-host verifies the live host before changing project authority.

Documentation

Project status

v0.2.0-beta.2 is a Beta for invited developers. It uses the Agent Plugins 1.0 format, with native CLI adapters for Codex and Claude Code on macOS and Windows, or a Linux memory host. Official VS Code integrations use the same adapter, but need a real graphical-chat acceptance check before they can be called supported: hook readiness and a manifest alone do not prove lifecycle delivery. Other agent clients remain disabled. See verification boundaries.

Licensed under Apache License 2.0.