Adapters connect to external data sources (REST API, PostgreSQL, Storage) to enrich AI answers via RAG. Assign categories to control which widgets use each adapter.
🔌 Adapters / Datasources
Name
Type
Categories
Status
Last Test
Actions
Loading…
Diagnostic
Query runs exactly what a chat request runs against this adapter, and shows the fragments it would contribute — without asking a model anything.
Resources lists what the adapter can see at all, which is the answer when it indexes nothing.
—
Each widget is an independent chat interface with its own categories, adapters, and visual config. Use the embed snippet to place widgets on any page.
💬 Chat Widgets
Name
Slug
Categories
RAG
Adapters
MCP
Public
Actions
Loading…
Conversations
What citizens typed into this widget. Erasing a conversation removes it from every table that holds it —
the messages, the queries and the telemetry rows behind them — not just the transcript.
0 keeps them until someone asks for erasure.
Session
Messages
Started
Last
Last message
Outgoing notifications. The platform posts a signed JSON body to your URL when something finishes. Delivery is queued, so a slow receiver never slows down the run that triggered it; a failed delivery is retried on a backoff and every attempt is in the log.
📡 Webhook endpoints
URL
Events
Signed
Status
Last delivery
Actions
Loading…
Deliveries
Event
Result
Code
Attempts
When
Error
RAG Profiles control reusable corpus folders, retrieval settings, preferred models, and one-click bootstrap, ingest, and training actions for every RAG-backed assistant.
RAG Profiles
Name
Category
Adapter
Folder
Model
Stack
Actions
Loading…
RAG Diagnostics
Profile
Corpus
Training
Widgets
Last Activity
Health
Loading…
🧠 Training Data
Category
Question
Answer
Score
Verified
Actions
Loading…
🎯 Fine-Tune Model
Training History
Date
Model
Samples
Status
Duration
Loading…
Fine-Tune Jobs (Editable)
Base
Output
Categories
Status
Actions
Loading…
📚 Knowledge Base
Title
Category
Source
Priority
Active
Actions
Loading…
🤖 Tenant AI Model Catalog
Per-tenant private catalog stored in tenant_ai_models. API keys are encrypted at rest and never echoed back to the UI.
Resolution order: widget → RAG profile → tenant default → env default → legacy provider.
Loading…
+ AI Model
Used by widgets/profiles via ai_model_id. Cannot be changed after creation.
⚙️ Legacy provider (fallback)
Used only when no catalog default is set, or when the catalog model fails.
Existing tenants keep working with no changes here.
🌱 Seed Default Data Sources
Creates 3 default adapters for this tenant: GIS Node API (localhost:8911/api/v1),
PostgreSQL (gis_test), Storage (H:\Dotwhite\Gis\gisnode\public\storage\).
Also creates default categories. Safe to run multiple times.
🗺️ MCP Servers
MCP servers expose resources (datasets, documents, layers) and
tools (read/search/query operations) to chat & RAG.
Servers can be in-process samples, your PostgreSQL/PostGIS database, a storage folder,
a REST API, or an external MCP HTTP server.
Loading…
—
+ MCP Server
Prefer api_key_env over inline secrets. Saved keys are masked on read.
👥 Users
Name / Username
Role
Status
Last Login
Actions
Loading…
🔧 Platform Settings
Key/value settings for this tenant. Use type json for complex values. Secrets are masked after save.
Loading…
🔍 Query Analytics
Use the AI Playground to test queries interactively.
Per-Class Confidence Thresholds
Raise the threshold for classes that get confused (e.g. car=0.50, truck=0.55).
Lower threshold = more detections but more false positives.
Overrides tenant default; widget override still wins.
Spatial coverage (optional)
Lets a regulation question answer from geometry — which zone polygon actually covers a
parcel — instead of only from the register's parcel reference. Both layers must be exposed
by the same connector. Leave empty and the answer keeps saying coverage was not checked.