Choose the AI arrangement that fits your group.

Why AI is used

Reading and comparing 70 responses, including revisions, is humanly difficult while a live discussion is moving forward. AI provides a fast first synthesis of themes, agreements, differences, unanswered questions, and distinctive ideas. The human facilitator reviews and edits the result before sharing it.

The AI is not the participant, grader, or decision-maker. It provides a fast first synthesis so the human facilitator can guide the next part of the conversation.

Privacy and responsible use

Before enabling synthesis, consider whether responses contain sensitive information, whether the selected model is permitted, how prompts and outputs are retained, and whether participants should be told that AI-assisted synthesis is being used. Policies differ by university and organization; LiveSynthesis supports responsible choices but does not determine whether a particular use is permitted.

Use an external AI tool without an API key

You do not have to configure a local Ollama model, manage an API key, or complete any LLM setup in LiveSynthesis. In the moderator History tab, Copy Anonymized Session Text opens a review window containing the session record with known participant identifiers replaced where possible and email-shaped identifiers removed. You can edit the copy, then paste it into ChatGPT, Gemini, Claude, or another approved AI tool that you already use.

This approach trades some convenience for flexibility: LiveSynthesis does not send the session to a provider, and the moderator chooses the external tool, model, and prompt. It also makes it possible to use an institution-approved model without configuring that provider inside LiveSynthesis. It is most practical for shorter sessions or when an approved external tool is required. With many participants or long responses, copying and pasting a large record into a general-purpose AI tool can be slow; the built-in workflow is designed to handle large numbers of participants and long responses.

Review the text before sharing it, remove anything sensitive that remains, and follow your university or organization’s policy for approved models, data handling, and participant disclosure. The resulting synthesis can be edited and pasted back into the session or shared with participants.

Use without AI

Questions, responses, messaging, attendance, and session history work without an LLM. Responses can still be reviewed, shared, copied, and exported. This option has no model cost and avoids sending response text to an AI service.

With AI: Cloud model

Cloud providers give access to frontier models and their latest capabilities without requiring the moderator to install or maintain a model. They can be useful for difficult, multilingual, or very large synthesis tasks.

The tradeoff is that response material leaves the moderator’s computer. Review the provider’s data-retention, training, security, and privacy policies, and manage the API key and spending limits carefully.

With AI: OpenRouter

OpenRouter provides one API key for access to a changing catalog that can include free, very low-cost, and frontier models. This makes it easier to compare quality and cost and to choose a less expensive model when the task does not require the most capable option.

OpenRouter is an additional service between LiveSynthesis and the selected model provider. Review OpenRouter’s terms and the selected model’s policies. Free models may have lower limits, less predictable availability, or lower synthesis quality.

With AI: Institution-approved model

A university- or organization-approved provider and model may offer safeguards for privacy, security, retention, access, and auditing that have already been reviewed. This can be the best fit when institutional approval is more important than using the newest or most capable model.

With AI: Local desktop model

A local desktop model provides complete privacy: no response data is sent outside the moderator’s desktop to a model provider. Local processing can be useful for sensitive or restricted discussions and typically has no per-request cloud charge.

The tradeoff is hardware. Ordinary computers may run small or medium models well, while very large models require more memory and processing power and may be slower or less capable than frontier cloud models.

With AI: Custom provider

The Custom provider option can connect to another compatible service, such as an organizational gateway, regional provider, or internal deployment. The moderator supplies the model-discovery and chat endpoints and an API key when required.

The organization is responsible for confirming the service’s privacy, security, retention, cost, and cross-origin requirements.

Costs and performance

Cloud cost depends on the number of participants, number of questions, length of responses, amount of session history included, model selected, and length of the synthesis. Provider prices and model availability change, so estimates should be dated and treated as approximate.

Performance includes more than model reputation. Consider synthesis quality, preservation of minority views, response time, language support, context capacity, reliability, and how well the result can support the next discussion.