Organization Settings
Manage organization details, members, and teams, configure AI providers and RAG behavior, and review usage, analytics, and activity logs all in one place.
Organization settings are the hub for running your workspace. Open them from Settings in the sidebar. Here you invite members and assign roles, register AI provider API keys, control how documents are processed (RAG), and review usage, analytics, and activity logs.
Most settings apply at the organization level and are shared with every member. What each person can change depends on their role.
Editing organization, AI, and RAG settings is limited to admins and above. Members and viewers can open the same screens but the save buttons and fields are disabled. Deleting the organization is reserved for the owner.
A tour of the settings
The left sidebar of the settings screen is split into three groups, each linking to the pages below.
Organization
Manage organization details (name, slug, default language), member invitations and roles, and team structure.
AI and RAG
Handle LLM, reranker, and web search API keys, plus embedding, captioning, speech transcription, chunking, and knowledge graph settings.
Monitoring
Review token usage and limits, pageview analytics, and the organization activity log.
Roles and permissions
Flux organization roles come in four tiers — viewer < member < admin < owner. Higher tiers carry more permissions. You assign a role when inviting a member, and an admin can change it later. The table below summarizes the role required for key tasks.
| Permission | Viewer | Member | Admin | Owner |
|---|---|---|---|---|
| View documents and agents | ||||
| Add and edit documents | ||||
| Delete documents | ||||
| Create and edit agents | ||||
| Invite members and manage roles | ||||
| Change organization, AI, and RAG settings | ||||
| View activity logs | ||||
| Delete the organization |
The owner is whoever first created the organization. Admins can do nearly all operational work except deleting the organization, which makes the admin role a good fit for delegating day-to-day administration.
Details, members, and teams
Refine your organization's basic details, invite the people you work with, and group members into teams.
Open the members screen
Go to Members in settings. The current member list and any pending invitations appear together.
Enter an email and role
Type the email address to invite and choose a role (admin, member, or viewer). An invitation email is sent automatically.
Manage invitations
You can resend or cancel pending invitations. Once the recipient accepts, they appear in the member list.
Organization details
Edit the organization name, the slug used in URLs, and the default language (Korean, English, or auto) under **Organization** in settings.
Teams
Group members into color-coded teams to keep things organized. Teams help you sort members neatly in larger organizations.
Deleting an organization cannot be undone. All documents, agents, conversations, embeddings, and knowledge graph data are permanently removed. Only the owner can do this, and you must type the organization name to confirm.
AI provider keys
When Flux calls an external AI service, it uses the API keys you register for the organization. Keys are stored encrypted, with one key per provider. Registered keys are then selected per node in the agent builder.
| Type | Supported providers | Used for |
|---|---|---|
LLM | OpenAI, Anthropic, Google AI, Groq, OpenRouter, vLLM (self-hosted) | Large language models for an agent's answer generation and reasoning |
Reranker | Cohere (the built-in Qwen3 reranker needs no key) | Reorders search results by relevance to improve retrieval quality |
Web search | Tavily, Serper, SerpAPI | Used by the web search tool node to pull live information from the web |
Open the providers screen
Go to Providers in settings. LLM, reranker, and web search appear as separate sections.
Add a key
Click Add, choose a provider, then enter a recognizable name and the API key. Self-hosted providers like vLLM also require a Base URL.
Confirm validation
On save, the key is validated automatically. A Validated badge means it works. If a key expires later, use the re-validate button to check it again.
API keys are stored encrypted and shown only partially masked. If a key is in use by an agent or another setting, you get a warning on delete and can force-delete only after seeing where it is used.
RAG settings
RAG settings decide how Flux processes and searches your documents after upload. Changes here apply to documents uploaded afterward.
Embedding models
Choose the text and image embedding models that turn documents into searchable vectors. You can use the built-in local models or an external provider.
Image and video captions
Automatically generate descriptions (captions) for images and videos so they become searchable. Pick the built-in vLLM or an external LLM and decide which formats to apply it to.
Speech transcription (STT)
Transcribe speech in audio files into text. Built-in Whisper is the default, and you can switch to an external transcription provider.
Chunking
Adjust the chunk size and overlap used when splitting documents into searchable pieces. The defaults suit most documents.
Knowledge graph (GraphRAG)
Configure the model and output language for the knowledge graph feature that extracts entities and relationships from documents.
Debug mode
When enabled, chat shows the retrieval and generation steps as a live timeline, which is helpful for inspecting how an agent builds its answers.
If you change the embedding model or dimension after documents are already uploaded, a migration is needed to reprocess existing documents with the new model. Follow the on-screen guidance. Search results may differ slightly while processing runs.
Usage, analytics, and logs
These screens show your organization's operational picture: how many tokens you spend, how the app is used, and who did what.
Usage
Review token usage and estimated cost over time, broken down by provider and model.
Quotas
Set monthly and daily token and budget limits along with warning thresholds. By default quotas are warning-only, so crossing a threshold sends an email warning. Turn on ‘Block requests when quota is exceeded’ and requests are blocked once the limit is reached.
Analytics
See pageview and event statistics and the most-visited paths at a glance.
Activity log
Records key activity in the organization such as member changes and setting edits in chronological order, with category filters.
Quotas are warning-only by default: crossing a threshold sends an email and does not block requests. However, if you turn on ‘Block requests when quota is exceeded’, requests are blocked once the limit (100%) is reached — choose the level that matches how strictly you want to control cost.
Frequently asked questions
The recipient must accept by clicking the link in the email to become a member. From pending invitations on the Members screen you can resend or cancel the invitation.
Admins and above can change each member's role on the Members screen. The owner's role, however, cannot be changed.
Embedding and reranking work with the built-in local models and need no key. LLM answer generation and web search usually require an API key for the relevant provider, though a self-hosted vLLM can connect with just a Base URL and no API key.
By default quotas are warning alerts, so crossing a threshold only sends an email and does not block anything. However, if you turn on ‘Block requests when quota is exceeded’, requests are blocked once the limit is reached.