A Complete AI Chatbot & Knowledge Base Platform
AI Assistant is a chatbot management platform built into the Noova admin. Each chatbot is provisioned as a Gradient agent endpoint with a selected LLM model, configurable parameters (maxTokens, temperature, topP, topK), and optional knowledge base attachments. Chat sessions support both streaming and non-streaming modes. The billing system tracks token usage per message with separate input/output pricing, charges wallets automatically, and supports grant-based quota distribution across teams. All access is controlled through 20 granular permissions covering chatbot and knowledge base operations.
Multi-Provider Model Selection
Select from 4 providers: OpenAI (GPT-4o, GPT-5, o1, o3), Anthropic (Claude Opus 4, Sonnet 4, Claude 3.5/3.7), DigitalOcean Gradient (Llama 3.3 70B, DeepSeek R1, Mistral Nemo), with model-specific pricing.
Document Embedding Pipeline
Knowledge bases use 3 embedding models (GTE Large EN v1.5, All MiniLM L6 v2, Multi QA MPNet Base) with async indexing jobs. Upload documents from DigitalOcean Spaces, track indexing status, and manage data sources.
Token Economy
VND-denominated wallets with per-1K token pricing for input and output separately. Usage logs track costs per chatbot and per user. Minimum balance floor (default 50 VND) prevents over-spending.
20 Granular Permissions
Chatbot permissions (View All/Shared/Own, Create, Edit, Delete, Use, Train, Analytics) and Knowledge Base permissions (View, Create, Edit, Delete, Upload, Analytics) - all role-based.
What AI Assistant Enables
Contextual Employee Support
Deploy chatbots grounded in your organization's knowledge base - company policies, HR procedures, training materials, and FAQs. Employees get answers sourced from your actual documents.
Course-Embedded AI Help
Attach chatbots to specific courses with course-scoped visibility. Learners get AI assistance within their learning context, with the chatbot trained on course-relevant knowledge bases.
Controlled AI Spending
Manage AI costs with token-based billing, wallet balance tracking, and grant quotas. Distribute budgets across teams with Equal Split or Shared Pool modes and per-member caps.
Model Flexibility
Choose the right model per chatbot - use GPT-4o for complex reasoning, Claude for nuanced responses, Llama for cost efficiency, or DeepSeek for specialized tasks. Switch models without rebuilding the chatbot.
Who Uses AI Assistant
System Administrators
Create and configure chatbots, manage knowledge bases, set up billing wallets, configure grant quotas, and control permissions. Monitor platform-wide usage and costs.
Training Managers
Deploy course-scoped chatbots that assist learners with course content. Attach relevant knowledge bases to provide contextual help within specific training programs.
HR Teams
Build chatbots trained on HR policies, benefits documentation, and procedures. Provide employees with instant answers to common HR questions from authenticated knowledge bases.
Finance & Operations
Monitor AI spending through wallet dashboards, set budget limits with grant quotas, track per-chatbot costs, and optimize model selection for cost efficiency.
How Organizations Use AI Assistant
HR Policy Knowledge Bot
An organization wanted employees to get instant answers about company policies, leave procedures, and benefits - without waiting for HR staff or searching through document folders.
- Created knowledge base with HR handbook, leave policy, benefits guide, and onboarding documents uploaded from DigitalOcean Spaces
- Deployed chatbot with Claude Sonnet 4 model for nuanced policy interpretation
- Set scope to 'public' so all employees can access from the platform
- Configured widget with company brand color and welcome message explaining available topics
- Token usage tracked per department through grant quotas with Shared Pool Per Cap mode
Course-Specific Learning Assistant
A training team needed AI assistants embedded in specific courses to help learners with questions about course materials, without exposing unrelated content.
- Created separate knowledge bases per course - each containing that course's lessons, reference materials, and FAQs
- Deployed chatbots with scope 'course' and linked to specific courseIds
- Selected GPT-4o mini for faster response times and lower per-token costs for high-volume student queries
- Enabled streaming chat so learners see responses generated in real-time
- Set display rules with priority ordering so course-specific bots appear before general ones
Budget-Controlled Department AI
An organization with multiple departments needed each department to have its own AI chatbot budget, preventing any single department from consuming disproportionate resources.
- Created wallet with organization-wide token balance funded through allocation orders
- Set up EQUAL_SPLIT grants distributing monthly token budgets equally across 5 department heads
- Each department head creates chatbots within their allocation - wallet charges tracked per user
- Finance monitors spending through usage logs showing cost breakdown by chatbot, model, and user
- Minimum balance floor prevents overspending - chat blocked when wallet drops below 50 VND threshold
Building an AI Chatbot in Four Steps
1
Create Chatbot and Select Model
Use the Chatbot Create Wizard to set up a new chatbot. Choose from 20+ LLM models across 4 providers: OpenAI (GPT-4o, GPT-5, o1, o3), Anthropic (Claude Opus 4, Sonnet 4, Haiku), Gradient (Llama 3.3, DeepSeek R1, Mistral Nemo). Configure model parameters: maxTokens (128-16,384), temperature (0-1), topP (0-1), and topK (0-10). Set chatbot type (system or user), scope (public, enrolled, or course), and attach to specific courses if needed.
2
Build and Attach Knowledge Base
Create knowledge bases and upload documents from DigitalOcean Spaces. The system indexes documents using embedding models (GTE Large EN v1.5, All MiniLM L6 v2, or Multi QA MPNet Base) through async indexing jobs. Track indexing status and manage data sources. Attach one or more knowledge bases to your chatbot via attachedKbIds - the chatbot will use these documents as context for generating grounded responses.
3
Configure Widget and Deploy
Customize the chatbot widget appearance: theme color, position on page, launcher shape, and welcome message. Set display rules that control when and where the chatbot appears - configure visibility conditions and priority ordering. The chatbot is provisioned as a Gradient agent endpoint and deployed as an embeddable widget within the Noova platform.
4
Manage Billing and Monitor Usage
Set up token wallets with VND balance. Configure grant quotas to distribute budgets across teams using Equal Split, Shared Pool, or Shared Pool Per Cap modes. Monitor usage through logs showing input tokens, output tokens, and costs per chatbot and per user. Track session history through the Chat Session Viewer. Adjust model selection, parameters, and knowledge bases based on usage patterns and cost optimization.
Complete Feature Breakdown
20+ LLM Models from 4 Providers
Select the optimal model for each chatbot based on capability, speed, and cost trade-offs.
- OpenAI - GPT-4o, GPT-4o mini, GPT-5, GPT-5 Nano, GPT-5 Mini, o1, o3, o3-mini
- Anthropic - Claude Opus 4, Claude Sonnet 4, Claude 3.7 Sonnet, Claude 3.5 Sonnet, Claude 3.5 Haiku, Claude 3 Opus
- DigitalOcean Gradient - Llama 3.3 70B, Llama 3.1 8B, DeepSeek R1 Distill Llama 70B, Mistral Nemo Instruct, GPT-oss 120B/20B
- Model parameters: maxTokens (128-16,384), temperature (0-1), topP (0-1), topK (0-10)
- Per-model pricing with separate input/output token rates
Knowledge Base with Document Embedding
Create knowledge bases that ground chatbot responses in your organization's actual documents.
- KB Types - public, site, course, documents - control scope and visibility
- Document upload from DigitalOcean Spaces - add files as data sources
- 3 embedding models - GTE Large EN v1.5, All MiniLM L6 v2, Multi QA MPNet Base Dot v1
- Async indexing jobs - background processing with status tracking (pending, processing, completed, failed)
- Pending ingest queue - batch document processing pipeline
- Attach multiple KBs to a single chatbot via attachedKbIds for combined context
Streaming & Non-Streaming Chat
Two chat modes to match different interaction patterns.
- Streaming - SSE-based real-time token generation. Messages update incrementally as the model generates
- Non-streaming - Standard request/response. Full answer returned at once
- Streaming flags - streaming (in progress), streamDone (completed), streamError (failed) on message documents
- Session management - conversations tracked in chat sessions with full message history
- Chat Session Viewer - admin interface to review past conversations
Token Billing & Wallet System
Metered AI usage with VND-denominated billing, wallet management, and cost tracking.
- Per-1K token pricing - input and output priced separately per model
- VND wallets - organization-wide token balance with debit/credit transactions
- Usage logs - track costs per chatbot, per user, and per model with timestamps
- Minimum balance floor - configurable minimum (default 50 VND) blocks chat when wallet is low
- Allocation orders - purchase token credits to fund wallets
- Pre-authorization - safety multiplier estimates max cost before streaming begins
Grant & Quota Management
Distribute AI budgets across teams with flexible allocation models.
- EQUAL_SPLIT - distribute budget equally across all members
- SHARED_POOL - centralized quota consumed by all members
- SHARED_POOL_PER_CAP - shared pool with per-member spending caps
- Member modes - SNAPSHOT (fixed member list at creation) or DYNAMIC (members added over time)
- Per-member tracking - individual usage within grants
- Grant lifecycle - create, activate, monitor, and close grants
Chatbot Configuration & Display
Control chatbot identity, appearance, and visibility rules.
- Chatbot types - System (platform-managed) and User (admin-created)
- Scope - public (all users), enrolled (enrolled users), course (specific course context)
- Course attachment - link chatbots to specific courses via courseIds array
- Widget appearance - theme color, position, launcher shape, welcome message
- Display rules - visibility conditions and priority ordering for multiple chatbots
- 20 permissions - Chatbot (View/Create/Edit/Delete/Use/Train/Analytics) and KB (View/Create/Edit/Delete/Upload/Analytics) with All/Shared/Own granularity
Frequently Asked Questions
Common questions about AI Assistant
Ready to Deploy AI Chatbots?
Start building AI chatbots with 20+ LLM models, document-grounded knowledge bases, streaming chat, token billing, and granular access control.
