Keys are stored locally on your machine, outside the project folder. They are never uploaded anywhere.
Enable LangSmith tracing to trace and debug agent runs. Leave blank to disable.
Semantic features add AI-powered metadata at ingest time. Disabled by default.
Use a different provider and model for concept extraction at ingest time. Leave on Default to use the same model as the agent.
ℹ Add models first via the Model Manager tab — they will appear here automatically.
Add model names per provider. They will appear in the Model dropdown in the sidebar.
Choose which of the models above each role may use for their own queries (see the Provider/Model selector in the sidebar). A role with nothing added here cannot query at all — superadmin always bypasses this and is never affected.
For Ollama, set this to whatever your own hardware actually supports (mirrors Ollama's own OLLAMA_NUM_PARALLEL) — this framework can't detect your VRAM. For cloud providers, this is a burst control on top of any RPM/TPM override below. Leave a provider unset for no cap.
Google gets auto-detected free-tier limits by default; every other provider tracks nothing until you add an override here. Set real numbers from your paid-tier account — leave a field blank to not track that dimension.
Override ingestion and search constants. Leave blank to use the compiled-in defaults. Changes take effect on the next ingest run.
Manage who can access this deployment and what they can do. Real enforcement always happens server-side — this panel is a convenience over the same API the CLI uses. A key created here is always an ordinary credential — whether its holder acts as a workspace admin or member is entirely determined by the grants below, not by anything set at key-creation time. Framework-level superadmin status can only be granted via the CLI (local-search auth create-key --superadmin), never from this panel.
| Workspace | Subject | Role | Granted by |
|---|
| Subject | Display name | Status | Created |
|---|