NextChat by Software Sushi

Independent seller packaging · NextChat 2.16.1 · Ubuntu 24.04 x86_64

Software Sushi supplies independent packaging and configuration. The seller logo does not imply upstream or AWS endorsement.

Run a single-instance NextChat web assistant on Ubuntu 24.04. Connect your own OpenAI-compatible chat provider after launch. NextChat 2.16.1 is supplied under the MIT license; Software Sushi charges for this repackaged software and its AMI configuration. EC2, EBS, network traffic and provider usage are charged separately. No provider subscription, credit, model weights or commercial NextChat service is included. Recommended instance: t3.medium, 4 GiB RAM and at least 24 GiB gp3 root storage. No GPU is needed for this web frontend. A model hosted on the same instance needs its own additional CPU, memory and storage.

Software price and separate charges

The software price is USD 0.03 per running instance-hour. If your Marketplace offer includes the five-day software trial for one concurrent instance, it covers that instance's software charge during the trial. Additional concurrent instances incur software charges. EC2, EBS, data transfer and model-provider charges still apply during the software trial. Check the actual offer for eligibility before subscribing.

Connect

Attach an EC2 role with Session Manager permissions, require IMDSv2, use encrypted storage, and permit outbound HTTPS to your provider and the regional Systems Manager endpoints. No inbound application rule is needed. Wait for cloud-init:

sudo cloud-init status --wait
sudo systemctl status sushi-nextchat-firstboot.service nextchat.service --no-pager

The web UI and its API listen at 127.0.0.1:3000 on the instance. From your workstation, with AWS CLI and the Session Manager plugin installed:

read -r -p 'EC2 instance ID: ' INSTANCE_ID
read -r -p 'AWS region: ' AWS_REGION
aws ssm start-session --region "$AWS_REGION" --target "$INSTANCE_ID" \
  --document-name AWS-StartPortForwardingSession \
  --parameters '{"portNumber":["3000"],"localPortNumber":["3000"]}'

Keep the tunnel open and visit http://127.0.0.1:3000. Use this same URL and browser profile each time. The first boot generates a different random access code for every instance. Retrieve it only in your own terminal, outside recordings:

sudo -iu ubuntu cat /home/ubuntu/.config/nextchat/access-code

Enter that code in NextChat's access-code screen/settings. The code protects model API requests; the static UI is visible through the local tunnel. This is one shared instance code, with no enterprise RBAC, per-user server accounts or shared server-side chat database. Restrict Session Manager access to trusted operators. Keep MCP disabled. The AMI does not expose a public HTTPS endpoint.

Configure a provider

A Session Manager shell may start as ssm-user. Before creating your provider file, switch to the application user:

sudo -iu ubuntu

Configure the provider on the instance using a file, so the key does not appear in shell history or process arguments. Run this as ubuntu in an interactive terminal. Enter a provider root URL, such as https://api.openai.com, without a trailing /v1. Enter an exact model ID available to your account.

python3 -c '
import getpass
import json
import os
from pathlib import Path

p = Path.home() / ".config/nextchat/provider.json"
value = {
    "base_url": input("Provider root URL: ").strip(),
    "model": input("Model ID: ").strip(),
    "api_key": getpass.getpass("Provider key: "),
}
fd = os.open(p, os.O_WRONLY | os.O_CREAT | os.O_EXCL | os.O_NOFOLLOW, 0o600)
with os.fdopen(fd, "w") as stream:
    os.fchmod(stream.fileno(), 0o600)
    json.dump(value, stream)
print("Saved credential file with mode 0600.")
'
sudo /opt/sushi/bin/sushi-nextchat configure \
  --provider-file /home/ubuntu/.config/nextchat/provider.json

The service restarts with the selected provider. Reload the browser, create a new chat, and choose the configured OpenAI model. User-supplied browser API-key overrides and configuration links are disabled. Keys up to 8192 characters are accepted; this configuration accepts one key without commas or whitespace.

An existing LocalAI or other OpenAI-compatible endpoint can be supplied by you. HTTP is accepted only for numeric loopback addresses, for example a separately configured service on 127.0.0.1; remote endpoints require HTTPS and valid TLS. This AMI neither installs nor modifies LocalAI. The endpoint must support /v1/chat/completions, streaming browser replies, and the configured model ID. For a loopback provider without authentication, enter its documented nonempty dummy token. Amazon Bedrock is not a direct OpenAI-compatible endpoint: use your own compatible gateway and its authentication, or another supported provider. No AWS model permissions are granted by this application.

For a later provider change, create a new mode-0600 file with the same fields and run configure with that file. Preserve the original file as your recovery copy. Do not paste keys in URLs, command arguments, screenshots or support logs.

Verify a real request

sudo -iu ubuntu bash /opt/sushi/docs/demo.sh

The demo sends a fresh synthetic inventory task through the running NextChat API to your configured provider. It requires a completed assistant response containing valid JSON, calculates the expected total independently, saves the input/request/ response and their hashes under /home/ubuntu/nextchat-demo/, and rereads the saved bytes before printing SUSHI_CUSTOMER_DEMO_PASSED. It consumes provider usage. Failed replies are retained in the run directory with mode 0600. This API demo does not add a conversation to the browser.

To reread a saved run without making another provider call, enter its directory:

read -r -p 'Full saved run directory: ' SAVED_RUN
/opt/sushi/bin/nextchat-demo --readback "$SAVED_RUN"

Conversations, backups and recovery

Conversations are stored in your browser's IndexedDB, with a localStorage fallback. Reloading the same origin/profile retains them, including when the instance service restarts. Clearing browser data, using another browser/profile, changing the tunnel's hostname or port, or browsing in a temporary profile can make those conversations unavailable. A server EBS snapshot does not back up browser chats. Use NextChat's export/backup controls and protect exported chats as sensitive files. Optional remote synchronization services are not configured or included.

The access code and provider configuration persist in /etc/nextchat/instance.json (root-owned, mode 0600). The firstboot service preserves them on subsequent boots. The matching operator access-code file is owned by ubuntu, mode 0600. Back up these files securely along with browser exports and any demo result directories. Never share configuration or browser exports in support tickets.

sudo /opt/sushi/bin/sushi-nextchat status
sudo journalctl -u nextchat.service --since '15 minutes ago' --no-pager
sudo systemctl restart nextchat.service

The launcher redacts configured codes and provider keys from application console logs; still review logs before sharing. A missing provider, invalid model, expired key or exhausted quota requires correcting your provider configuration. The UI can start before a provider is configured; model requests cannot succeed then. For a missing access-code copy, run sudo systemctl restart sushi-nextchat-firstboot.service to restore it from the existing instance configuration. If the root configuration is lost or damaged, restore your secure backup before restarting the service; do not delete it as a troubleshooting shortcut. Restore exact files and permissions, and run firstboot again to refresh the operator copy.

The container runs as UID/GID 1000 with a read-only application filesystem and temporary cache. Software Sushi supplies the startup and credential/log-redaction adapter around the original upstream image. Provider credentials are configured after launch and are not bundled. The provider is configured on the instance, not through browser credential overrides. The per-instance access code is the browser credential for model requests; it is not an individual user account. Trusted application code can read its live credentials, so this is not isolation from malicious application code running under the same user.

Additional original dependency notices and source archives are under /opt/sushi/licenses/nextchat/supplemental/, in its notices/ and archives/ folders. See /opt/sushi/licenses/nextchat/THIRD-PARTY-NOTICES.md for Edge Runtime's MPL source locations and the attribution of browser compatibility data to caniuse.com under CC BY 4.0. Those components retain their own terms. The included version is the upstream stable release from July 2025; provider protocol changes can require an updated AMI. This package includes the self-hosted web application only, without a promise of upstream hosted or commercial features.