Get started with NextChat¶
Launch the published image, open its private browser interface and configure one model provider. The walkthrough uses your provider and can incur usage charges.
1. Launch and connect¶
Subscribe through the Marketplace offer
and launch package 0.1.2 on t3.medium or a larger listed type. Use at least
24 GiB encrypted gp3 root storage and require IMDSv2. Attach an EC2 role with
Session Manager permissions and allow its required outbound HTTPS connectivity.
Keep inbound application ports closed. Allow outbound HTTPS to your chosen provider.
Open a Session Manager shell and wait for setup:
sudo cloud-init status --wait
sudo systemctl status sushi-nextchat-firstboot.service nextchat.service --no-pager
2. Open the browser interface¶
On your workstation, install the AWS CLI and Session Manager plugin and sign in with your own authorized AWS identity. Start the tunnel and leave it open:
read -r -p 'EC2 instance ID: ' nextchat_instance_id
read -r -p 'AWS region: ' nextchat_region
aws ssm start-session --region "$nextchat_region" --target "$nextchat_instance_id" \
--document-name AWS-StartPortForwardingSession \
--parameters '{"portNumber":["3000"],"localPortNumber":["3000"]}'
Open http://127.0.0.1:3000. Use the same URL and browser profile on later visits so your locally stored conversations remain visible.
In your private instance terminal, outside recordings, retrieve the access code:
Enter it in NextChat's access-code screen or settings. Treat it as a secret.
3. Configure a model provider¶
Follow Provider configuration to create
a private file and apply it. Use your provider's root URL without /v1 and an
exact supported model ID. After configuration, reload the browser, start a new
chat and select the configured OpenAI model. Browser API-key overrides are disabled.
Provider data and charges
Your endpoint receives the prompts you send. Check its data policy, model access and budget first. The AMI does not create a provider account or supply credits. An AWS Bedrock endpoint needs your own compatible gateway.
4. Run the saved-result walkthrough¶
From the instance shell:
This makes a provider request with synthetic data, saves the input, request,
response and receipt under /home/ubuntu/nextchat-demo/, and independently checks
the saved result. Successful completion prints SUSHI_CUSTOMER_DEMO_PASSED.
It does not create a conversation in the browser or establish model quality.
To reread a saved run without another provider request, switch to Ubuntu's account:
Then provide the full saved directory printed by the walkthrough:
read -r -p 'Full saved run directory: ' nextchat_saved_run
/opt/sushi/bin/nextchat-demo --readback "$nextchat_saved_run"
Continue with backups and service management.
Last updated November 16, 2026 · Suggest an improvement