Console

The Console is the admin dashboard for the EgoX Orchestrator. It's where a project owner configures a project, wires up tools and knowledge, and monitors what production traffic is doing. It is not an end-user chat app — the /ask API is the product; the Console is the control plane.

Console is	Console is not
Admin dashboard for project owners	End-user chat interface
Configuration management	Production traffic handler
Usage analytics & auditing	Customer-facing product
Testing / debugging playground	Real-time conversation app

The user journey

1. Create an account
2. Create a Project (Tenant)
3. Configure it:
     • pick model & temperature
     • write system prompts
     • add tool endpoints (APIs the AI can call)
     • upload knowledge documents (RAG)
4. Test it in the Playground
5. Grab credentials           (Tenant ID + API key)
6. Integrate /ask in your own backend
7. Production traffic flows:   Your App → /ask → EgoX Orchestrator
8. Monitor usage & cost in Analytics
9. Audit conversations in Threads

Areas of the Console

The Console has a global area (all your projects) and a project area (one project's config + monitoring).

Global

Area	Purpose
Dashboard	Overview across all projects — total tools, documents, threads, API calls; recent activity; token-usage trend.
Projects	Create, list, edit, and delete projects (tenants).
Account Settings	Profile, preferences, avatar.
Access Tokens	Generate / list / extend / revoke API keys for the `/ask` API.

Inside a project

Grouped as Configure → Monitor → Test:

Group	Area	What it does
Configure	AI Settings	Model, temperature, history length, feature toggles (Tools / RAG), response format.
Configure	Prompts	Edit system prompts per intent: `base`, `rag`, `tool`, `rag_tool`. Variable highlighting (`{{context}}`, `{{query}}`).
Configure	Endpoints	Define REST/GraphQL tools the AI can call — schema, auth, retries, forwarded headers.
Configure	Knowledge	Upload documents (PDF/TXT/MD); watch chunking + embedding; preview chunks; test similarity.
Monitor	Analytics	Token usage, estimated cost, API calls — over time, by model, by intent. CSV export.
Monitor	Threads	Browse and audit every conversation; drill into a thread for turns, tool calls, and token counts.
Test	Playground	Test the configuration before production — chat UI + debug panel (intent, tool calls, RAG chunks, raw request/response).

Configuring a project

AI Settings

Model — e.g. gpt-4.1-mini (default), gpt-4.1, gpt-4o.
Temperature — deterministic ↔ creative.
History length — how many prior messages to include as context.
Features — toggle Tools and RAG on/off.
Response format — text by default; request json_object + a JSON schema per call.

Prompts

One system prompt per intent, so the AI behaves correctly for each request shape. The four tabs map directly to the four intents: Base (vanilla), RAG (rag), Tool (tools), RAG+Tool (rag_tools).

Endpoints (Tools)

Each tool is an external API the LLM may call during /ask. Configure:

Input schema + example input/output — what the LLM sends.
Static headers — always-sent headers (e.g. an upstream API key).
Forwarded headers — an allowlist of inbound /ask headers to pass through (e.g. x-locale). Sensitive headers are blocked server-side.
Auth type — FORWARD passes the caller's bearer token; NONE sends no auth.
Retries — storm-protected (dedup, per-tool cap, transient retry, iteration cap).

Tools can also be created by AI agents over MCP — those are tagged source = 'mcp' so you can review them as drafts before they go live.

Knowledge Base (RAG)

Drag-and-drop documents; EgoX chunks and embeds them into the EgoX DB. Watch processing status, preview chunks, and test which chunks a query retrieves.

Monitoring & auditing

Analytics — headline tiles (Total Tokens, Est. Cost, API Calls) plus token-over-time, usage-by-model, and usage-by-intent breakdowns. Export to CSV.
Threads — a searchable, paginated audit log of every conversation. Open a thread for a read-only, turn-by-turn view including tool calls, arguments, responses, intent, model, and per-message token counts.

Design & stack

Aesthetic — an "AI command center": deep navy/charcoal backgrounds, electric-cyan + purple accents, monospace for data, subtle glow/particle animations.
UX principles — progressive disclosure, real-time feedback, contextual help, keyboard-first (Cmd+K palette).
Stack — React 18 + TypeScript, Vite, React Router v6, Zustand, TanStack Query, Tailwind CSS, Framer Motion, Radix UI, Monaco Editor, React Hook Form + Zod.

Integrate the API → Client SDK
Manage tools/knowledge from an AI agent → MCP connection

On this page