Does it work offline?

The gateway and the local model can run offline on the device, but chat platforms such as WeChat or Telegram need the internet to deliver messages to it. The cloud route also needs access to the AI service you configure.

Is conversation content sent to third parties?

With a local model, message content is not sent to any AI provider, though it still passes through the chat platform you connect. With a cloud model, message content goes to the AI provider you configure.

How good and how fast is the local model?

Local options are Qwen3.5 2B (about 1.9GB quantized) and 4B; 2B is faster and 4B slower. Small models suit everyday questions; for complex reasoning, connect a cloud model or switch between the two per use.

Which chat platforms are supported, and how do I connect one?

More than 20, including WeChat, Telegram, Discord, Slack, DingTalk and Feishu. Add a platform under Channels in the gateway page, enter the bot token or webhook it asks for, authorize, then send a test message.

Can I change the model?

Yes. Add OpenAI, DeepSeek or any compatible AI service under Settings > Models and switch any time, with no change on the platforms. The local model size is chosen at deployment.

Why can't I open the management page?

The page accepts local connections only by default. For a remote device, forward port 18789 to your computer over SSH, open that local port, and sign in with the Gateway Token from the deployment log.

What should we check on licensing before commercial use?

OpenClaw is an open-source project on GitHub, used under its repository licence; the local models keep the Qwen publisher's terms; and each chat platform's bot rules apply. Check each before shipping a closed product.

OpenClaw AI Gateway

Deploy OpenClaw AI messaging gateway on reComputer, with optional local AI model for privacy-first conversations.

Beginner15minOn-Device LMSource Code

reComputer J4012, the host for the AI compute gateway route that runs the model locally

Reusable Site Templates

Places where staff or customers should ask an AI inside the chat apps they already use, without a separate integration per platform.

A shared team assistant — one AI assistant in a Slack, Discord, DingTalk or Feishu group for the whole team.
A personal assistant in WeChat or Telegram — chat with models such as DeepSeek or OpenAI right in the messaging app.
Conversations that must not reach an AI provider — run the model locally on a Jetson, so message content is not sent to a third-party AI.

Not a fit: work that needs large-model reasoning quality, since the local route runs 2B-4B models. The cloud route sends message content to the AI provider you configure.

How it works

OpenClaw device composition: chat platform messages reach a reComputer J40 Series or reComputer R1100 Series gateway host; the R1100 connects to a cloud AI service, and the J40 optionally does too

Three pieces: the chat platforms you already use, a reComputer running the gateway, and the AI model that answers. The gateway passes each message to the model and sends the reply back to the same platform. The model can run on the same Jetson or be a cloud service.

WeChat, Telegram, Discord, Slack, DingTalk, Feishu and more than 20 platforms in all, added one by one in the gateway's web page.

Credentials once per platform: enter the bot token or webhook the page asks for, then authorize.
One integration to maintain: changing the model or adding a platform happens in the gateway, not on each platform.

The two routes differ in where the model runs; the gateway software is the same.

Route	Host	Model	Resources
AI compute gateway	reComputer J40 Series	Local model on the host GPU (Qwen3.5 2B / 4B), cloud models optional	12GB disk, 8GB RAM
Lightweight gateway	reComputer R1100 Series	Cloud AI services	4GB disk, 4GB RAM

The gateway's management page is on port 18789 and accepts local connections only by default.

Access	How
On the host	Attach a display and open port 18789 locally in a browser
From your computer	Forward port 18789 over an SSH tunnel, then sign in with the Gateway Token from the deployment log

How well it works

The table lists what each route provides, and its limits.

Route	What it does	Limit
AI compute gateway (reComputer J40 Series)	A local model answers; message content is not sent to an AI provider; no token fees	Local models are 2B-4B, and 4B is slower than 2B
Lightweight gateway (reComputer R1100 Series)	Connects cloud models such as OpenAI or DeepSeek, switchable any time	Reply speed depends on the cloud service; pay per call; message content goes to the cloud

Both routes connect more than 20 chat platforms at once.

What you get out of it

Once deployed, these are the entry points.

Entry	What it does	Notes
Each chat platform	Users send messages; AI replies land in the same platform	Added under Channels in the gateway page
HTTP 18789	Gateway management page: test chat, platform connections, model settings	Bound to localhost by default; remote access via SSH tunnel and Gateway Token
HTTP 11434	Local model service	Present only when the compute gateway enables a local model

The Gateway Token is kept in a config file under the deploying user's home, readable only by that user.

How to deploy

Tell us about your site — we work out the hardware, then you connect the data and follow the steps.

Message Data

Can message content be sent to a cloud AI provider? The gateway software is the same either way. This decides where the model that answers runs.

No — it must stay on the device

A Qwen3.5 2B or 4B model runs on the host GPU

Yes — pay-per-call cloud models are fine

OpenAI, DeepSeek and other cloud models over a stable network

Some conversations can, others must stay local

A local model plus cloud models added in the gateway settings, switched per use

Deployment Guide · 15min →

FAQ

For cloud models, a reComputer R1100 Series; the gateway needs only 4GB of disk and 4GB of RAM. For a local model, a reComputer J40 Series, which then needs 12GB of disk and 8GB of RAM. Both need Docker installed first.

Didn't find your answer?

We Are Glad to Be Your Hardware Partner !