Web Platform
The browser-based interface for managing every aspect of your Upivia workspace. Dashboard for overview, Agents Hub for agent lifecycle, Agent Studio for deep configuration, AI Builder for terminal commands, and the Operate section for approvals, activity, and tasks.
Dashboard
The Dashboard is the first page you see after signing in. It gives you an at-a-glance overview of your workspace's health and activity.
- Balance — Current org wallet balance with a low-balance warning banner when funds drop below the threshold. Click to go to Billing and top up via Stripe.
- Monthly spend — Total spend this month across all services, broken down by agent and operation. Helps you track burn rate and spot anomalies.
- Active agents — Count of agents currently in active status. Click to go directly to the Agents Hub.
- Pending approvals — Service requests awaiting human review. Shows count and links to the Approvals tab in Tasks Hub.
- Recent activity — Latest 10 service requests with status (executed/blocked/approval_required/failed), cost, and timestamp. Each row is expandable for full payload details.
- Worker budget — For team members: your personal budget allocation, current spend, and remaining balance. For admins/owners: org-level summary.
The dashboard uses a responsive layout with SVG sparkline charts for spend trends, skeleton loading states, and real-time polling for live updates. Hover over any sparkline point to see exact spend for that period.
Agents Hub
The Agents Hub is the central page for managing your agent fleet. It consolidates four previously-separate pages into one tabbed interface.
| Tab | Content |
|---|---|
| Sessions | ChatGPT-style chat interface. Left sidebar shows live TUI sessions from the CLI (green dot = online, gray = offline) and chat history. Right panel is full-height scrollable chat with streaming responses, voice input via microphone (browser SpeechRecognition), and voice output via Kokoro TTS. Agent picker dropdown to switch which agent you're chatting with. Model selector to choose the LLM. Create new sessions, switch between them, and view history — all sessions are synced with the server and appear in the AI Builder terminal too. |
| My Agents | Responsive grid of all your agents. Each card shows the agent's icon (editable by admins via the Lucide icon picker), name, key mode badge (test/live), visibility badge (private/team/org), status (active/paused/disabled), category label, service count, and monthly budget summary. Filter the grid by team and category using the CategoryBar pills. Click any agent card to expand the Agent Studio with 5 sub-tabs. Click the + tile to create a new agent. |
| Automations | Three sub-tabs: Workflows — build DAG automations with the visual workflow editor (10 step types, drag-and-drop or JSON). Scheduler — create once or recurring scheduled tasks with cron expressions, view upcoming runs in a calendar, monitor task stats. Published — browse agents published to your team or organization with visibility badges and unpublish controls. |
| Integrations | Two sub-tabs: Connected Apps — OAuth connections to Slack, Gmail (Nylas), HubSpot, Salesforce, Jira, and Linear. Each connection shows status, provider, account label, and install/revoke controls. Custom APIs — create your own HTTP services with method, URL, headers, auth, pricing, and timeout. Published custom APIs appear alongside built-in services in the catalog. |
Sessions Tab
ChatGPT-style chat, voice I/O, live TUI session indicators. Switch agents and models per session. All sessions synced with the server.
Agent Studio
Click any agent card in the My Agents grid to expand the Agent Studio. It provides five sub-tabs for deep configuration of a single agent. Changes are saved instantly via the PATCH API and reflected across all views.
| Tab | Features |
|---|---|
| Identity | View and edit: name, description, category, and icon (Lucide icon picker with search). Agent Soul — six rich personality presets with backstory, behavioral rules, and communication style. Presets: Research Analyst (academic, citation-focused), Code Reviewer (line-by-line review with severity labels), Customer Support (empathetic, plain language), Data Scientist (rigorous methodology), Concise Builder (no-nonsense, code-first), Creative Strategist (multiple approaches, metaphors). Max 2000 chars. System Prompt override for full control. Voice Configuration — Kokoro TTS: enable/disable, rate slider (0.5–2.0), pitch slider (0–2), language (e.g. en-US), Test Voice button. |
| Services | Three sections. Skills — workflow capabilities + enabled services as a unified skill list, self-adjusting proposals from workflow agents (approve/reject), create from templates (Research & Summarize, Fetch & Notify, Data Processing, Daily Report). Service Bindings — two-column layout: Enabled (left) | Available (right). Search bar and category filter pills. Per-service card with toggle, expandable body showing: operations with pricing, monthly budget cap ($), daily request limit (count), approval threshold ($), per-operation payload defaults (dynamic form from JSON schema). Delegations — compose new delegations (select child agent, set budget cap, write objective, grant permission subset), track inbound and outbound tasks with status and budget progress bars. |
| Memory | Three.js WebGL 3D force-directed graph. Node types (card types): knowledge (blue glow), experience (green), insight (purple), plan (orange), constraint (red). Edge types: supports (green curved line), satisfies (blue), constrains (orange), conflicts (red) — with directional light pulses traveling along edges. Interactions: drag to rotate, scroll to zoom, click node to focus with animated camera transition and glow flash. Sidebar with filter pills by card type and clickable node list. Detail panel shows node title, outcome sign, and all connections. Memory composer: title, content, scope (agent/team/workspace). Semantic search with badges. Card actions: archive, delete. |
| Activity | Agent accountability dashboard. Chain integrity banner: green 'Audit Chain Intact' or red 'Audit Chain Broken' with reason. Six summary stat cards: Total Events, Service Calls, Delegations, Total Spend ($), Errors, Blocked. Three views: Timeline (chronological events with expandable JSON payloads, event type icons, hash-verified link icon, mode badge), Delegations (inbound/outbound with status, objective, budget/spent, model used), Service Calls (sortable table with operation name, status, cost, mode). Cursor-based pagination with Load More. |
| Settings | Publishing — request publication to team or org scope, view pending requests and active publications with scope badges, unpublish. Clone Agent — duplicate with all service bindings, budget caps, and settings; new API key generated. Delete Agent — type-to-confirm gate prevents accidental deletion. YUPI and chat agents are protected from deletion. |
Identity Tab
Edit name, description, icon. Choose from 6 personality presets or write a custom system prompt. Configure Kokoro TTS voice.
AI Builder Terminal
The AI Builder is a fixed-position panel at the bottom of every page (hidden on the Sessions tab to avoid redundancy). It's a terminal-style interface for managing your entire agent fleet through commands or natural language conversations with YUPI. Modelled after VS Code's integrated terminal with frosted glass translucency.
- Three modes — Plan (propose only, no execution), Build (emit commands for auto-execution), Ask (Q&A, no commands).
- Model picker — DeepSeek V3, GPT-4o Mini, GPT-4o, Claude 3.5 Sonnet, Gemini 2.0 Flash.
- Session management — Multiple named sessions with create/delete/switch. Auto-synced with the server — sessions appear in the Sessions tab. Delete also closes the server-side chat session.
- 40+ direct commands — create, delete, rename, clone, pause, resume, rotate-key, enable, disable, budget, config, publish, unpublish, spawn, delegate, memory add/delete, schedule, workflow create/steps/run/grant, swap-yupi.
- Auto-approval toggle — When enabled, AI-suggested commands are queued for confirmation. Press Enter to run all or type 'n' to skip. When disabled, commands run immediately.
- Voice input/output — Microphone button for speech-to-text (browser SpeechRecognition). TTS toggle reads assistant responses aloud via Kokoro.
- Dock position — Bottom (default, spans full width) or Right (docks to right side, full height). Resizable via drag handle (80–600px bottom, 280–800px right). Collapsible to 36px header bar.
- swap-yupi <agent> — Switch which agent powers the AI Builder. Persists to localStorage across page reloads.
- Frosted glass — Translucent panel with 20px blur and saturation boost. Theme-aware (adapts to light/dark mode).
Tasks Hub & Activity
Tasks Hub (/tasks)
The Tasks Hub consolidates four previously-separate pages into one tabbed interface for operational oversight. Each tab has its own polling, filtering, and action controls.
| Tab | Description |
|---|---|
| Approvals | Lists all pending service request approvals. Each row shows the agent, operation, estimated cost, and time pending. One-click approve or reject. Admins only — non-admins see an empty state. Approvals expire after 24 hours. |
| Budget Requests | Workers request budget increases from team managers. Shows requester, current allocation, requested amount, reason, and status (pending/approved/rejected). Managers can approve with a custom amount or reject with a note. History view shows all past requests. |
| Publishing | Review inbox for publish requests (agents, workflows, APIs). Admins see all pending requests across the org. Approve or reject with a reason — the creator gets notified via inbox. Approved publications are visible per their scope (team/org). |
| Agent Requests | Agent-to-human communication. Agents can request clarification, budget increases, plan approval, secrets, or report blocked actions. All org members can view and respond. Used by the TUI's local_ask_user tool and agent-requests API. |
Activity Hub (/activity)
Full visibility into your workspace's operations with two tabs:
| Tab | Description |
|---|---|
| Usage | Usage records filtered by agent, service, status, mode (test/live), and date range. Shows cost per operation with model and token info. SVG sparkline chart for spend trends. Per-agent and per-period breakdowns. |
| Audit Logs | Every service request, approval, publish event, and memory mutation is recorded here. Expandable JSON payloads for full detail. Filter by event type. Hash-chain verification with integrity banner. Export at /settings/danger. |
Inbox (/inbox)
Unified inbox with client-side interactivity: Notifications (publish decisions, system messages — mark as read with optimistic UI update), Invitations (accept or decline workspace invites with one click), and Workspaces (list all workspaces you belong to, leave with sole-owner guardrails that prevent orphaning an org). Filter pills: All, Unread, Notifications, Invites, Workspaces. Pagination with Load More.