Skip to content

Tabby Coding Assistant

AI Agents Tabby v0.32.0 Hardened AMI

Tabby Coding Assistant is an Ubuntu 24.04 AMI that runs Tabby, the open-source, self-hosted AI coding assistant server for code completion and chat. IDE extensions send code context to the server, and it answers with completions and chat replies from a model you control. The server runs on CPU, listens on loopback only, and starts without any model weights: you attach the model you choose after launch.

Get started View on AWS Marketplace
AMI release
0.1.3 (Tabby v0.32.0)
Recommended
t3.medium
Storage
16 GiB+ encrypted gp3
Access
Session Manager, no inbound ports
Released
Oct 12, 2026
Delivery
Amazon Machine Image in your own AWS account

Self-hosted AI code completion and chat server, so code context stays in your AWS account.

Starts with no model downloads. Attach the completion model you choose using the included example config.

The server binds to localhost and you connect through Session Manager. No inbound ports are needed.

Quick start

Full guide →
  1. Subscribe
  2. Launch the instance
  3. Connect
  4. Run the walkthrough
  5. Check health yourself
  6. Reach the API from your computer

What is included

Component Details
Operating system Ubuntu 24.04 LTS, x86_64, CPU only
Tabby v0.32.0 CPU server, /opt/tabby/bin/tabby, with its llama-server runtime for local models
Service tabby.service, running as the tabby system user with state under /var/lib/tabby
Embedding stub sushi-tabby-embed.service, a small loopback stand-in for an embedding model, so Tabby starts without downloading one
Model example /opt/sushi/etc/tabby/customer-model.example.toml, with commented blocks for a local completion model and remote chat and embedding backends
Self-checks A walkthrough at /opt/sushi/docs/demo.sh and a self-test at /opt/sushi/bin/smoke.sh

No completion, chat or embedding model weights are included. Completions and chat start working once you attach a model. The Tabby Enterprise web UI is turned off (--no-webserver), and GPU and CUDA builds are not included.

Ports

All listeners are on 127.0.0.1. Nothing listens on a public interface, and you do not need any inbound security group rule.

Address What it is
127.0.0.1:8080 Tabby HTTP API, including /v1/health
127.0.0.1:18080 Loopback embedding stub used by the default configuration

Reach the API from your computer with Session Manager port forwarding. Tabby on this image has no authentication in front of it, so do not open port 8080 in a security group.

  • t3.medium (2 vCPU, 4 GiB RAM) for the server without local models.
  • At least 16 GiB encrypted gp3 root volume.
  • IMDSv2, and outbound HTTPS.

Local models need more RAM and disk. Choose a larger instance and volume before you download weights.

Billing

You pay an hourly software fee for each running instance, shown on the AWS Marketplace listing. An optional five-day free trial covers the software fee for one instance and then converts to hourly billing. It is the same software before and after the trial. EC2, EBS, data transfer and any model provider's charges are billed separately.

Where to go next

  • Getting started: launch, connect, check health and reach the API from your computer.
  • Configuration: attach a completion, chat or embedding model.
  • Operations: security model, service management, logs, updates, backups and support.
  • Troubleshooting: common problems and fixes.

Upstream project

Tabby is published by TabbyML. Content outside the ee/ directory is under the Apache-2.0 license; the ee/ tree uses the Tabby Enterprise license and is not part of this offer. Software Sushi packages Tabby independently and does not claim TabbyML endorsement.

Last updated October 13, 2026 · Suggest an improvement