Documentation▾

Synology AI Console and Office

Connect Office's AI Assistant to Porten. Choose a model below to prepare the values you need in Synology; the settings update from the current catalog.

Connect in Synology

  1. In AI Console, add an integration named porten.ai and choose OpenAI-compatible API.
  2. Set Base URL to https://porten.ai/v1. Enter your existing Porten API key directly in Synology.
  3. Click Pull Generative Model List and choose the same model as above. Confirm Context window against the model's current serving capacity.
  4. Set Output tokens to your chosen budget, run Synology's verification and save the integration.
  5. Select the integration for Synology Office. Try AI Assistant in a new document with synthetic text before using your regular documents.

If the model selector is unavailable

Use the model catalog, or query GET https://porten.ai/v1/models with your API key from your own client. Choose a model that supports chat and use its exact id.

The top-level context_length describes the reported input-plus-output window. For an unloaded model, it may be the catalog ceiling rather than a verified serving window. Load the model in your client, refresh the list and verify the context before saving. If the context is missing, do not guess it from the model name. The public selector does not include private models or your key's restrictions; Synology's authenticated model list is the final check.

Context and output are different settings

Context window is shared by your input and the generated output. Documents, conversation history, instructions and formatting all use space. A model may support a larger theoretical window than the nodes currently serve.

Output tokens is your cap on the generated response. Reasoning also counts toward this cap, even when Office does not display it. A higher cap can allow a longer response, but does not guarantee a better or faster answer.

For example, a 32768-token serving window cannot fit 30000 input tokens and 4096 output tokens. Reserving the entire window for output leaves no room for your input. See models and serving context and the API reference.

Choose a budget for your work

4096 is a tested starting budget for short Office tasks with Qwen3.6-27B, not a universal minimum or model limit. On 9 November 2026, Office completed a short email rewrite and a meeting summary at context 32768 / output 4096, preserving dates and amounts. Other models and longer documents need their own validation.

  • Check that representative answers are complete and useful. A short visible answer can still use many reasoning tokens.
  • For truncated or empty responses, try a larger output budget, such as 8192, while leaving enough room for the input.
  • For context_length_exceeded, shorten the document or history, reduce the output budget, or choose a model with a larger verified serving window.
  • In API responses, finish_reason: "length" means a length limit was reached. Synology may not expose this field, so also inspect the answer itself.

Porten does not impose a universal output default when a client omits the limit. The value you enter in Synology is the client's budget.

📄 Reading as a machine? This page is available as raw Markdown at https://porten.ai/docs/synology.md — or grab the whole site via llms.txt / llms-full.txt.