Xeretron AI Server

Apps · Control Plane · v2.45

Agent Platform

Not a chatbot.
The AI agent Control Plane.

A centralized AI agent platform with permissions, knowledge (RAG), memory, teams, tools, vision browser, human approvals, visual automations, and live observability — governed by an Orchestrator, not the UI.

  • Create, version, and run agents
  • RAG, memory, and team delegation
  • Human approvals and least privilege
  • Visual automations + webhooks
  • xrtk_… APIs for CRM / ERP / bots
  • Windows · UI EN / ES
Request access See modules

For operations with governed internal agents, API integrators, and multi-workspace orgs that need isolation and audit. Requires the Xeretron ecosystem (AI Server · Sync · Orchestrator).

Agent Platform: agent control plane with chat and tools
Click to enlarge

Capabilities

End-to-end governance

Versioned agents

Types: chat, document, task, and autonomous. Every edit creates a snapshot. Clone, activate, deactivate — no hardcoded features.

Knowledge RAG

Local bases: upload or URL, indexing, and links to agents. Retrieval before the LLM — if it fails, the turn continues.

Human approvals

Human gate per tool. Pending queue, history, and floating panel. Least privilege: the model only sees granted tools.

Teams and delegation

Delegates in the same workspace. One hop. The lead coordinates; the delegate can require approval.

Automations

Manual, schedule, webhook, or event triggers. Visual node editor (n8n-style). Runs and logs.

Live view

Agent presence, active chats, browser, delegations, and events. Real-time observability.

Control Center · 14 modules

The full control plane in one app

Dashboard and Live view

Workspace KPIs, composition, pipeline, and recent activity. Live: presence, active chats, browser, delegations, and floating hub.

Agents, conversations, and tasks

CRUD, clone, multi-tab editor. Streaming chat with VL and attachments. Task queue with prepare → execute → finalize timeline.

Automations and approvals

Visual node editor, triggers, and webhooks. Human gate with pending queue and history.

Knowledge, models, and tools

RAG bases, model catalog ≤35B (VL unlocks images), read-only builtins, and restricted Python scripts.

APIs, reports, and settings

xrtk_… keys, Orchestrator/Sync URLs, and in-app manual. CSV / Excel / TXT export. Language, timezone, and AI Server link.

Agents

The heart: dynamic configuration

Not fixed features. Versioned entities: identity, instructions, models, tools, knowledge, team, memory, plugins, and permissions.

  • Types: chat · document · task · autonomous
  • Every edit = snapshot + version
  • Grants with “requires human approval”
  • Session or persistent memory
  • Internet Browser plugin (VL vision + forms)

Turn pipeline

  1. User message
  2. RAG if knowledge bases exist
  3. Streaming LLM
  4. Tool loop: grant → approval → execute
  5. Reply (attachments / browser when needed)

Flutter never orchestrates or enforces permissions. The Orchestrator does.

Plugin · Internet Browser

Browser with vision and forms

On Windows with WebView2: navigate, tabs, click, type, scroll, screenshot, extract_text, submit, and cookies. VL flow: viewport capture → pixel click → extract / forms.

Up to 3 tabs

Browser panel in the conversation; background sessions.

VL vision

Best with multimodal models: the agent sees and acts on the page.

Useful SERP

Structured organic results for research and workflows.

How it works

Control Center + Orchestrator + Xeretron

Control Center

Desktop app: create, operate, observe. Never runs tools or enforces permissions on its own.

Orchestrator

Source of truth: agents, grants, RAG, memory, approvals, tasks, automations, and live.

AI Server · Sync · Files

Streaming inference (models ≤35B), login with link_code, OCR/extract for chat and knowledge.

Isolated workspaces and role memberships. Talk about organizational isolation — not cloud “projects” as a product UI feature.

Security and governance

Backend enforce-all. UI = hints.

Encrypted keys

xrtk_… API keys and credentials encrypted at rest. Generate, reveal, revoke.

FS isolation

Per-agent data. Scripts without free imports; I/O only under the agent root.

Audit

Auth, agent CRUD, tools, approvals, models, and nodes — operational traceability.

Integrations

APIs for CRM, ERP, and bots

Same Bearer on REST and WebSocket. In-app integration manual. Automation webhooks. Exportable reports by agent and date range.

  • Platform API keys xrtk_…
  • Workspace live channel
  • CSV / Excel / TXT export
  • Sync login: email + password + XXXX-XXXX code
14UI modules
4agent types
≤35Bmodels
2.45version

Platform

Windows

Desktop Control Center client. Browser plugin with WebView2. UI in English and Spanish. macOS coming soon.

Windows

FAQ

Frequently asked questions

Is it a chatbot?

No. It’s the Control Plane: create, manage, run, and orchestrate agents with permissions, RAG, memory, tools, and automations.

Do I need Xeretron?

Yes. AI Server for inference, Sync for login/link_code, Orchestrator as backend, and File Server for OCR/extract when needed.

Are there cloud projects?

The real model is Organization → Workspace → memberships/roles. Isolated workspaces — not a “projects” product UI feature.

Which platforms are supported?

For now Windows only. macOS is coming later.

Who enforces permissions?

The Orchestrator. The UI shows hints; the backend enforce-all.

Next step

The agent Control Plane of the Xeretron ecosystem.

Govern agents with permissions, knowledge, teams, and automations — version 2.45.

Talk to the team See all apps