Administration
Organization settings
Choose the models behind Workers, Agent Mode, titles and background analysis, set sync and analysis defaults, and add your own model API keys.
La documentation existe en anglais et en turc, comme l'interface de Harmona. Cette page est affichée en anglais.
Admins find these under Management → Organization Settings. General holds the defaults for the whole organization. API Keys holds your own model provider keys. Changes on General take effect when you select Save settings.
Choosing a model
Each model setting has a provider list and a model list. Each model shows its relative cost, and Harmona's recommended default is marked; Restore default brings it back. Options with a key icon use your organization's own API key.
| Setting | What it controls |
|---|---|
| Worker Settings → Model | The model that powers Workers. |
| Agent Mode → Model | The model that powers Agent Mode (Create with Harmona). Shown only where Agent Mode is available. |
| Chat Settings → Conversation titles | Names new conversations after their first reply, in chat, Agent Mode and Workers. On by default. The title model writes the titles; the default is a small utility model. |
Connection settings
| Setting | Default | What it does |
|---|---|---|
| Sync interval | 1 day | How often syncing connections (SFTP, SharePoint, Google Drive, ReadMe, Web Crawl) re-sync. Set it in minutes, hours or days. |
| Auto image recognition | On | Uploaded images are analyzed by a vision model. When off, images are stored unanalyzed and can be analyzed later from the file's detail page. |
| Auto deep document analysis | Off | Every document (PDF, Word…) and analyzed image uploaded from now on is transcribed into a table agents can compute over. One AI call per file. Files already uploaded aren't analyzed; use Analyze on a file's detail page. |
| Automatic document entity indexing | Off | Each new document gets a small card of what it is about (cities, products, counterparties). The cards answer "all X" and "X grouped by Y" questions completely. When off, the index updates only when an admin runs Generate cards from a connection's Deep Analysis panel. |
The two analysis settings each have their own model. The deep-analysis model also handles files you analyze by hand. For entity indexing, a small, fast model is recommended; the agent's own model still writes the answers.
Analysis models
- CSV & Excel analysis model: reads uploaded spreadsheets and notes what each column means, so agents query them accurately.
- SQL analysis model: analyzes connected database tables and writes their business descriptions (Analyze Tables).
- SAP HANA analysis model: the same for SAP HANA tables, which are often larger. Worth raising if your SAP tables are big.
A setting you never change keeps following Harmona's default.
Model API keys (bring your own key)
Under Organization Settings → API Keys you can add keys for OpenAI, Anthropic, Google and Azure OpenAI. That provider's models then appear with a key icon, both on General and in agent settings.
- 1Select Add API Key.
- 2Choose the provider and paste the key. A key name is optional.
- 3Fill in the optional provider fields if you need them: Organization ID for OpenAI, Project ID for Google, or API base URL, API version and deployment name for Azure OpenAI.
- 4Select Create.
Mis à jour 2026-09-24