Insights & summaries
AI & models has separate pages for Insights and Summaries. Both reuse credentials from AI Providers. The old /admin/settings/ai/assistants address redirects to Summaries.
Insights agent
Section titled “Insights agent”Open AI & models → Insights (/admin/settings/ai/insights). This page and the conversational Insights feature require Team or Enterprise, including when you supply your own API key.
Enable Insights agent, choose a Provider, set its Model, and click Save. It answers questions about recorded usage, cost, and activity. In the tenant-wide view it can also query event data through KurrentDB Flight SQL, which requires licensed KurrentDB 26.1 or later.
| Configuration key | Default | Meaning |
|---|---|---|
Insights:Agent:Enabled | off | Enables the conversational agent. |
Insights:Agent:Provider | Anthropic | Anthropic, OpenAI, Kurrent AI, Cohere, or OpenRouter. |
Insights:Agent:Model | selected vendor’s judge default | Model for a provider using your key. Kurrent AI uses its deployment-managed model. |
Insights:Agent:MaxRows | 5000 | Query result row limit; accepts 1–100000. |
Insights:Agent:TimeoutSeconds | 30 | Query timeout in seconds; accepts 5–600. |
The Insights:Agent:* keys replace the older Analytics:Agent:* names. Legacy values remain fallbacks when the corresponding new key is unset.
See Insights for using the conversational feature.
Session titles
Section titled “Session titles”Open AI & models → Summaries (/admin/settings/ai/summaries). The Session titles panel controls server-generated titles, including the refresh action beside a session title.
Choose Auto or a provider. Auto considers configured providers that have a usable model: it prefers a single tenant-supplied Anthropic, OpenAI, or OpenRouter key, otherwise managed Kurrent AI, then Cohere. If several tenant providers remain and neither managed fallback is available, choose a provider explicitly.
The provider key is SessionTitle:Provider. Run Timeout (SessionTitle:RunTimeoutSeconds) sets each provider attempt’s budget, from 5–120 seconds, default 30. Save after changing these settings.
Models
Section titled “Models”| Key | Built-in default |
|---|---|
SessionTitle:Anthropic:Model | claude-haiku-4-5 |
SessionTitle:OpenAI:Model | None: enter a model explicitly. |
SessionTitle:Cohere:Model | command-a-03-2025 |
SessionTitle:OpenRouter:Model | openai/gpt-4o-mini |
Kurrent AI’s model is deployment-managed. A blank OpenAI model makes that provider unavailable for title generation even when its API key is set. In Auto mode the page shows model sections for usable configured vendors.
Auto-backfill
Section titled “Auto-backfill”The Auto-backfill switch periodically titles ended sessions that still have a raw first-message title. It is off by default.
| Key | Range | Default |
|---|---|---|
SessionTitle:AutoBackfill:Enabled | on/off | off |
SessionTitle:AutoBackfill:IntervalSeconds | 60–3600 seconds | 300 |
SessionTitle:AutoBackfill:MinEndedAgeMinutes | 1–60 minutes | 2 |
SessionTitle:AutoBackfill:MaxAttempts | 1–10 | 3 |
SessionTitle:AutoBackfill:BatchSize | 1–100 | 20 |
Imported sessions have a separate title-completion path, enabled by default: an import that arrives with only a raw title can be titled by the server even when this general backfill switch is off. This completes missing import titles; it does not retitle every imported session.
What’s Done
Section titled “What’s Done”The What’s Done panel configures the outcome summary shown on ended sessions and its classification. Choose Same as session title to inherit the title provider, or select a provider explicitly.
WhatsDone:Provider controls this selection. Model overrides use WhatsDone:Anthropic:Model, WhatsDone:OpenAI:Model, WhatsDone:Cohere:Model, and WhatsDone:OpenRouter:Model. Their built-in defaults match the title defaults above; OpenAI again requires an explicit model. Kurrent AI’s model is deployment-managed. Click Save to apply changes.