Coding Instances

What a coding instance is

A coding instance is a shared AI coding server. It runs a frontier-class open-source coding model on a dedicated GPU and exposes it over an OpenAI-compatible endpoint (and the Anthropic-style /v1/messages path). Your squad points Cursor, Continue, aider or any SDK at it by changing one URL, and codes without per-token metering or rate limits.

It is meant for developer squads and small teams, up to 10 members on one instance. If you need a GPU for an application instead, such as a chatbot, OCR or embeddings, create an inference instance.

How it works

  1. Choose a coding model. Larger models write better code and need more GPU. The wizard shows what fits and what it costs. See Supported Models.
  2. Choose how it should feel. Pick a speed preference (Patient Thinker, Flow State or Nitro) and how much context the instance should hold. Both set the GPU requirement. See Quants and Context.
  3. Set the squad size and a weekly GPU budget. The tier caps how many GPU hours the instance can run per week: 10, 20 or 40.
  4. Add credit. You need at least $5. Credit is deducted every 15 minutes while the instance runs.
  5. Provisioning. syndicAI finds a dedicated GPU, deploys the model and brings the instance up, usually in 2 to 5 minutes.
  6. Invite your squad. Each member joins with an invite link and creates their own API key, so usage is tracked per person.

Who pays

The owner's credit pays for the instance. The owner can cover it alone or split it with the squad equally. See Billing and Pricing.

Running on rented GPUs

Instances run on rented spot GPUs. The instance starts and stops with use to save GPU hours, and if a host is lost syndicAI provisions a new one and brings the instance back. See How It Works for the lifecycle.

What happens to your code

syndicAI does not store or log your prompts, code or the model's responses. Requests are relayed through the syndicAI proxy on Cloudflare to your GPU, which runs the model. The proxy reads a request only to enforce size limits and to detect image content, and keeps nothing. The GPU is dedicated to your instance and not shared with other customers. For dedicated nodes inside a specific country, see syndicAI Enterprise.

Where to go next