This guide covers the LLM settings page which is part of the Discourse AI plugin.
Required user level: Administrator
The dedicated settings page is designed to have everything related to Large Language Models (LLMs) used for Discourse AI features in one place.
Depending on the Discourse AI feature enabled, an LLM might be needed. Please check each Discourse AI feature to know if an LLM is a pre-requisite.
Features
- Add new models, with prepopulated information
- Add custom models not mentioned
- Configure LLM settings
- Allow specific LLM use for AI Bot
- See the AI Bot username
- Enable vision support (model dependent)
- Configure allowed attachment types
- Set up per-group usage quotas
- Track input/output token costs
- Test
- Save settings
Adding LLM connections
- Go to
Admin→Plugins→AI - Go to the
LLMstab - Add a new connection, pick your model
- Add in the API key (depending on the model, you might have more fields to input manually) and save
- (Optional) Test your connection to make sure it’s working
Supported LLMs
You can always add a custom option if you don’t see your model listed. Supported models are continually added. Pre-configured models are templates — you can always achieve the same result using “Manual configuration”.
Anthropic
- Claude Opus 4.6
- Claude Sonnet 4.6
- Claude Haiku 4.5
- Gemini 3 Pro
- Gemini 3 Flash
OpenAI
- GPT-5.4
- GPT-5 Mini
- GPT-5 Nano
Open Router
- DeepSeek V3.2
- Moonshot Kimi K2.5
- xAI Grok 4 Fast
- MiniMax M2.5
- Z-AI GLM-5
- … and many many more
Additionally, hosted customers can use the CDCK Hosted Small LLM pre-configured in the settings page. This is an open-weights LLM hosted by Discourse, ready for use to power AI features.
Configurations fields
You will only see the fields relevant to your selected LLM provider. Please double-check any of the pre-populated fields with the appropriate provider, such as
Model name
Core fields:
-
Display name— the friendly name shown in dropdowns -
Model name— the model identifier sent to the API (e.g.claude-sonnet-4-6,gpt-5.2) -
Provider— the service hosting the model (e.g. Anthropic, OpenAI, Google, AWS Bedrock, Azure, Open Router, etc.) -
URL— the API endpoint URL (not shown for AWS Bedrock) -
API Key— configured via the AI Secrets system Tokenizer-
Max prompt tokens— controls prompt trimming to prevent oversized requests Max output tokens-
Input cost/Output cost— cost per million tokens, used for usage tracking -
Cached input cost/Cache write cost— for providers that support prompt caching -
Vision enabled— enables image understanding (model dependent) -
Allowed attachment types— file types the model can process
Provider-specific fields (shown dynamically based on selected provider):
- AWS Bedrock:
Access Key ID,Role ARN,Region, reasoning/thinking options,Prompt caching - Anthropic: reasoning options,
Prompt caching - OpenAI:
Organization ID,Reasoning effort,Service tier - Google:
Enable thinking,Thinking level - Open Router:
Provider order,Provider quantizations
Quotas (available after initial save):
- Per-group usage quotas can be configured with max tokens, max usages, and duration
Technical FAQ
What is tokenizer?
- The tokenizer translates strings into tokens, which is what a model uses to understand the input.
What number should I use for Max prompt tokens ?
- A good rule of thumb is 50% of the model context window, which is the sum of how many tokens you send and how many tokens they generate. If the prompt gets too big, the request will fail. That number is used to trim the prompt and prevent that from happening
Caveats
- Sometimes you may not see the model you wanted to use listed. While you can add them manually, we will support popular models as they come out.
Last edited by @sam 2026-03-24T04:55:48Z
Check document
Perform check on document: