System Health

Checking…

API Docs

API Endpoints

GET /api/v1/health

POST /api/v1/chat

GET /api/v1/corrections

GET /api/v1/admin/databases

POST /api/v1/admin/databases

PUT /api/v1/admin/databases/{id}

POST /api/v1/admin/databases/{id}/test

POST /api/v1/admin/databases/{id}/schema

POST /api/v1/admin/databases/{id}/embed

DELETE /api/v1/admin/databases/{id}

Swagger: /docs · ReDoc: /redoc

Tenants

Customer organisations that own one or more databases. Each tenant has a business name, the domain where their UI is hosted, and an admin contact.

New Tenant

Business Name *

Domain

Status

Max Client Databases (entitlement — baked into a signed token so the tenant cannot bypass it)

Admin Contact

Name

Contact Notes

Chat UI branding (display name, accent color, sample prompts) and end-user auth (Auth0, hash) are configured by the tenant admin at their own site via the local-admin tool. Acufy does not hold those values.

Target Databases (shadow)

Acufy-side shadow records for tenant target databases. No credentials live here — the tenant stores those in their own registry at their site. Each shadow carries an auth key the tenant must present when provisioning the target DB locally.

Filter

Issue License Key

Tenant *

Host *

Port

Database Name *

Industry

Display Name

Description

Re-issue will revoke the current license key and mint a fresh one. The tenant will need to re-register.

License Key Issued

Share the key below with the tenant admin out-of-band (e.g. email or secure message). They must paste it verbatim into their local registration tool.

Tenant Code

Target

License Key

This value is not re-displayed on subsequent views. If lost, re-derive via re-issue.

—

Total

—

Active

—

Inactive

—

Vector Enabled

Client Databases

Starting… 0%

System Knowledge Console

Ask about system capabilities, tables, rules, and configuration

Ask about system capabilities (e.g., "what do you know about trading partners?") or run data queries (e.g., "list all active members").

Setup Registry Database

Enter the PostgreSQL server credentials below. The application will create the registry database (if it does not already exist) and initialise all required tables.

Host

Port

Username

Password (required)

Use AWS IAM Authentication (RDS / Aurora)

IAM auth generates a short-lived token at connection time. SSL is enforced automatically.

AWS Region *

AWS Access Key ID

AWS Secret Access Key

AWS Session Token (optional)

Database Name

The database will be created if it does not already exist.

Working…

SQL Generation Hints

Domain-specific hints injected into SQL prompts to improve query accuracy.

Generate Hints connects to the live database to sample actual coded column values and row counts, then sends that data alongside the schema to Claude. The AI produces hints grounded in real values — not guesses — covering coded columns, business rules, preferred join paths, and table notes. All active hints are injected into every SQL generation prompt for this database. Pull the schema first if you haven't already.

Manual Code Column Labelling For columns missed by Generate Hints — add authoritative labels manually

Generate Hints already handles common coded columns automatically (any column named *_status, *_type, *_code, *_flag, *_cd, etc.). Use this scanner only for columns with non-standard names that were missed, or when you want to provide exact authoritative labels rather than relying on Claude's inference. Scans all text columns with ≤10 distinct values.

Select a database above and click Scan Now to discover code columns.

Select a database to view its hints.

Entity Intelligence Prompts

Per-tenant system prompts used when the chatbot narrates an entity snapshot (member, group, plan, provider, claim, or any custom type). When no active prompt exists for an entity, the orchestrator falls back to the legacy hardcoded prompt.

Database

Select a database to view its intelligence prompts.

Entity Detection Patterns

Python regex patterns the chatbot uses to recognise entity IDs in user questions (e.g. member IDs, NPI, claim IDs). Multiple patterns per entity_type are allowed — lowest priority wins. When no active pattern exists for a given entity_type, the detector falls back to the hardcoded healthcare defaults.

Database

Select a database to view its entity patterns.

New Entity Pattern

Database *

Entity Type *

Recognised by the built-in dispatch: member, family, group, plan, npi, claim.

Regex Pattern *

Python regex syntax. Must have at least one capturing group — the first non-empty group is the entity ID.

Trigger Keywords (comma-separated)

Priority (lower = applied first)

Description (optional)

Active (uncheck to disable without deleting)

Single-Token Router Configuration

When a user enters a single bare token (e.g. "1234", "ORD987"), the chatbot can run a deterministic SQL lookup against the configured view and bypass the LLM. At most one active config per tenant. Leave unconfigured to disable the router for that tenant.

Database

Select a database to view or create its single-token config.

Schema Pruner Configuration

Anchor tables are always included in the pruned schema regardless of the question. Use for central entities the LLM always needs visibility into.
Forced companions are bridge/child tables that must travel with a trigger table (when the trigger is selected, its companions are pulled in too).
When a tenant has no config row, the pruner runs purely score-driven — no tables are forced in.

Database

Select a database to view or create its pruner config.

New Intelligence Prompt

Database *

Entity Type *

Prompt Text *

Active (uncheck to disable without deleting)

Query Cache Warming

Generate sample questions and pre-populate the AI cache so users get instant responses

Cache Warming Pre-populate the AI cache so users get instant responses

Database

Questions to generate

Generating…

Enter one question per line. Click Load Questions to add them to the list below.

Review & Edit Questions

0 / 0 0%

Warmed Query Cache

Permanently stored question→SQL pairs (PostgreSQL). Survive server restarts. Time-relative questions are excluded.

Select a database above to view its warmed cache entries.

Live Query Cache

Exact-match cached answers (Redis / in-memory). Each entry is evicted automatically when its TTL expires.

Select a database above to view its live cached queries.

Failed Query Repair

Queries that exhausted all strategies are listed below. Re-run them after adding domain hints or corrections to fix them automatically.

Pending: 0

Fixed: 0

Select a database above to scan for failed queries.

Last Repair Results

Correction Memory Validation

Test each stored correct_sql against the live client DB. Identifies corrections that are now broken due to schema changes.

Provider Settings

Configure per-client LLM and embedding providers. Leave fields blank to inherit system defaults from .env.

This client is using system defaults from .env. Fill in fields below and save to override per-client.

Select a database above to configure its provider settings.

Fast LLM

SQL generation · result narration · RAG answers

Provider

Model

API Key · key saved

Endpoint URL (Azure / Ollama / self-hosted)

Reasoning LLM

Validation · hint generation · query suggestions

Provider

Model

API Key · key saved

Endpoint URL (Azure / self-hosted)

Embedding Provider

Correction memory · vector index · semantic cache

Changing provider requires Re-Embed

Provider

Model

API Key · key saved

Endpoint URL (self-hosted TEI)

Vector Dimension

Similarity Threshold (0–1, default 0.82)

Connection Test Results

User Feedback

Bad result reports submitted from the user UI.

Loading…

Feedback Detail

Error Rate Trend

Daily fallback % and strategy breakdown.

Daily Fallback %

Date	Total	SQL OK	SQL Error	Vector	Fallback	Fallback %
Loading…

Failing Questions

Most-repeated questions that triggered fallback strategy.

Loading…

LLM Cost Tracking

Daily token consumption per client.

Daily Total Tokens

Date	Queries	Input Tokens	Output Tokens	Total Tokens	Avg / Query
Loading…

Query Review

Browse all user queries. Click any row to open the Correction Workbench.

Time	Question	Strategy	Status	Time (ms)	Actions
Loading...

Created	Question	Reason	Uses	Actions
Loading...

Retrain System

Re-pull schema, regenerate embeddings and hints, clean up stale corrections, and flush all caches. Run this after any database schema changes or when the system needs a fresh start.

Database

Skip hint generation (faster — use if hints are already good)

Progress

Running

Correction Detail

Question

Original (Bad) SQL

Corrected SQL

Correction Workbench

Question

SQL

Ask AI to Fix

Describe what's wrong — the AI will regenerate the SQL using your feedback.

Results

Click "Run SQL" to see results

Approve & Save Correction

Once you're satisfied with the SQL, save it as a correction. Future similar questions will use this SQL automatically.

Resolve Review

This query was flagged for admin review. Choose a resolution:

Resolved

Add Cache Entry

Question (parameters extracted automatically)

SQL (paste or generate above)

Add Hint

Hint Type

Table (optional)

Column (optional)

Hint Text

Synonymous Terms (comma-separated)

All these user-facing words will map to the same database field.

Maps to (table.column or description)

Generated Hint Text Preview

System Health

API Endpoints

Tenants

New Tenant

Admin Contact

Target Databases (shadow)

Issue License Key

License Key Issued

Client Databases

Setup Registry Database

SQL Generation Hints

Entity Intelligence Prompts

Entity Detection Patterns

New Entity Pattern

Single-Token Router Configuration

Schema Pruner Configuration

New Intelligence Prompt

Query Cache Warming

Failed Query Repair

Last Repair Results

Correction Memory Validation

Provider Settings

Connection Test Results

User Feedback

Feedback Detail

Error Rate Trend

Daily Fallback %

Failing Questions

LLM Cost Tracking

Daily Total Tokens

Query Review

Retrain System

Progress

Correction Detail

Correction Workbench

Add Cache Entry

Add Hint

Register Database

License

Option 1 — Dev Mode Auto-Issue

Option 2 — Paste Acufy-Issued Token

Danger Zone

Schema

Admin Sign In