Recommended deployment for self-hosted OpenHuman Core with remote desktop clients

[Synap-project]

Recommended architecture for self-hosted OpenHuman Core with remote desktop clients

Hi everyone,

First of all, congratulations on the project. I'm really impressed with OpenHuman and I'm currently evaluating it as the cognitive layer of a local-first AI platform.

My goal is to centralize everything on a dedicated Ubuntu server:

• OpenHuman Core
• Ollama
• Memory
• Embeddings
• Workspace
• Local data

The desktop application would simply act as a remote client, similar to how VS Code connects to a remote machine.

The architecture I'm aiming for is:

macOS Desktop App
        │
        │ Remote connection
        ▼
Ubuntu Server
├── OpenHuman Core
├── Ollama
├── Memory
├── Embeddings
├── Workspace
└── Local data

Today I installed OpenHuman_0.58.7_amd64.deb on Ubuntu 24.04. The installation completed successfully, but I realized that this package is the desktop application, which requires an X11/Wayland display.

After reading the Cloud Deploy documentation, I understand that OpenHuman Core can expose a JSON-RPC endpoint (OPENHUMAN_CORE_RPC_URL and OPENHUMAN_CORE_TOKEN), so I would like to confirm the intended deployment model.

Questions

1. Is there an official way to install OpenHuman Core only on an Ubuntu server without the desktop UI?

2. Can the macOS desktop application connect to a self-hosted OpenHuman Core running on another machine?

3. Is this the recommended production architecture, or is the desktop application currently expected to run on the same machine as the Core?

4. If this architecture is supported, what is the recommended installation process?

If possible, I'd like to keep all memory, embeddings, workspaces and AI models centralized on the server, while allowing one or more desktop clients to connect remotely.

Thank you for your time, and congratulations again for such an interesting project.

[Synap-project]

Update

After some more investigation on Ubuntu 24.04, I wanted to share what we've found.

Initially we installed:

• OpenHuman_0.58.7_amd64.deb

This installs the desktop application, which is expected to require an X11/Wayland display.

Later we downloaded:

• openhuman-core-0.58.7-x86_64-unknown-linux-gnu.tar.gz

This package contains the standalone openhuman-core binary.

Running:

./openhuman-core --help

shows that it provides a full CLI, including commands such as:

• run
• doctor
• memory
• workspace
• config
• voice
• skills
• agent
• workflows

We also ran:

./openhuman-core doctor report

which reports that the runtime expects a structure under:

~/.openhuman/users/local/

including:

• config.toml
• workspace/
• memory_tree/
• daemon_state.json

So it looks like the desktop application and OpenHuman Core are indeed separate components.

Our remaining question is whether the intended production deployment is:

• OpenHuman Core running permanently on a Linux server
• one or more desktop applications connecting remotely to that Core

or whether the desktop application is still expected to run locally alongside the Core.

We're really enjoying exploring the project and would love to understand the recommended architecture before integrating it into our environment.

Thanks!

[Synap-project]

Update: confirmed local Core architecture

After some more investigation, we've been able to validate a significant part of the architecture.

Our setup:

• Ubuntu 24.04
• openhuman-core 0.58.7
• Headless installation (no desktop environment)

Confirmed

• openhuman-core runs perfectly without a graphical environment.
• The default HTTP server listens on 127.0.0.1:7788.
• /health and /schema are publicly accessible.
• /schema exposes the complete JSON-RPC API (around 700 methods).
• /rpc is protected with Bearer authentication.

We also confirmed that the local authentication token is stored in:

~/.openhuman/core.token

Using that token:

TOKEN=$(cat ~/.openhuman/core.token)

curl -X POST http://127.0.0.1:7788/rpc \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"core.version","params":{}}'

returns:

{
  "jsonrpc":"2.0",
  "id":1,
  "result":{
    "version":"0.58.7"
  }
}

So we've confirmed that third-party software can communicate directly with OpenHuman Core through its JSON-RPC interface.

The only remaining question is whether the Desktop application is already capable of connecting to an existing remote OpenHuman Core, or whether that capability is still planned.

Hopefully these findings are useful to anyone trying to deploy OpenHuman as a server component.

[Synap-project]

Further investigation completed.

We've successfully deployed OpenHuman Core in a completely headless Ubuntu environment and confirmed that it can be consumed as a standalone service.

Highlights:

• Core runs without any graphical environment.
• HTTP API available on localhost:7788.
• /health and /schema are public.
• /rpc uses JSON-RPC 2.0 with Bearer authentication.
• The local authentication token is stored in ~/.openhuman/core.token.
• We successfully called:
  • core.ping
  • core.version
  • openhuman.config_get

The configuration API exposes the complete runtime configuration, including action directories, memory backend, autonomy settings, local AI configuration, model registry, etc.

At this point we've confirmed that third-party software can integrate directly with OpenHuman Core through its documented JSON-RPC interface.

The only remaining question is still whether the Desktop application can connect to an already-running remote OpenHuman Core.