Skip to main content
The MITHUNAI Web Console provides an integrated interface for managing organizational knowledge, configuring multi-agent assistants, designing embeddable widgets, and inspecting groundedness telemetry. Every action taken in the console executes against the underlying /arukz/api/v1 HTTP API, ensuring identical behavior across graphical and programmatic workflows.

​
1. Ask & Playground

The Ask view is an interactive sandbox for querying assistants and verifying citation grounding in real time:
  • Token Streaming: Responses stream with Server-Sent Events (SSE), delivering immediate visual feedback.
  • Interactive Citation Badges: Inline numeric badges ([1], [2]) expand on hover or click to reveal the exact document title, source URL, and quoted passage.
  • Evidentiary Inspection: When an assistant abstains, the playground displays the confidence evaluation breakdown and explains which required facts were absent from the knowledge base.
  • Conversation History: Previous inquiry threads are persisted and searchable in the left sidebar.

​
2. Knowledge Manager

The Knowledge section organizes enterprise content into modular collections:
  • Collections View: Group documents by product line, security classification, or audience (e.g., Public Docs, Internal SRE Runbooks).
  • Source Connectors: Add new sources via graphical modals for GitHub, GitLab, Sitemaps, Notion, and File Uploads.
  • Sync Telemetry: View live synchronization status badges (indexing, completed, failed), along with ingested document counts, chunk quantities, and byte sizes.
  • Chunk Inspector: Drill down into individual documents to view how Markdown AST and source code was segmented into vector embeddings.

​
3. Assistant Studio

The Assistants view configures how models synthesize information:
  • Model Switcher: Select from available LLMs (Claude 3.5 Sonnet, GPT-4o, Gemini 2.0 Flash) or connect private on-premises models.
  • Knowledge Attachment: Toggle checkboxes to bind one or more knowledge collections to the assistant.
  • Instruction Prompt Editor: Customize the assistant’s persona, technical level, and formatting instructions.
  • Temperature & Guardrails: Calibrate temperature (default: 0.0 for maximum factual consistency) and enforce strict Answer-First grounding constraints.

​
4. Deployments & Channels

The Channels view provides distribution controls for exposing assistants to users:

​
Widget Designer

  • Live Preview Canvas: Test the interactive chat bubble in a simulated desktop or mobile viewport.
  • Brand Theming: Customize accent colors, light/dark themes, greeting messages, and placeholder prompt chips without touching CSS.
  • Embed Generator: Copy the one-line HTML snippet containing your deployment key (arukz_wk_...).
  • Allowed Origins: Enter domain names (e.g., https://docs.acme.com) authorized to render the widget.

​
API Keys & MCP Server

  • Service API Keys: Generate SHA-256 hashed secret keys (arukz_sk_...) with designated roles and custom rate limits.
  • Instant Revocation: Revoke compromised keys with immediate zero-propagation delay.
  • MCP Config Generator: Generate ready-to-paste JSON configuration blocks for Claude Desktop and Cursor.

​
5. Analytics & Gap Discovery

The Analytics view monitors how users interact with your assistants:
  • Query Volume & Latency: Charts displaying total inquiries, cache hit ratios, and p95 response latencies.
  • Abstention Rate: Track the percentage of questions where the assistant declined to answer due to missing facts.
  • Unanswered Questions Triage: Review a curated list of questions that could not be answered, enabling technical writers to identify and draft missing documentation.

​
6. Organization Settings

The Settings view manages enterprise administrative controls:
  • Team Members: Add team members by email address and assign roles (Admin, Editor, Member).
  • White-Label Branding: Upload custom SVG logos, set organization display names, and configure favicon assets.
  • Support Access: Explicitly grant temporary, audited read access to MITHUNAI support engineers for troubleshooting.
Last modified on September 27, 2026