AverQel OS
Technical Guidebook

AverQel Documentation

The in-app source of truth for AverQel: grounded chat, DeepSpace, documents, memory, connectors and MCP, providers, privacy, and product architecture.

What AverQel Does Now

AverQel combines grounded retrieval, DeepSpace chat, provider routing, connector and MCP integrations, persistent memory, and a working document editor.

The command surface is chat. DeepSpace keeps the conversation, provider selection, safe tool execution, memory, and assistant-message persistence tenant-scoped and separate from the MCP runtime.

Durable runtime readiness

DeepSpace keeps chat history and assistant answers durable so a reload can restore the conversation without requiring an IDE-style workspace or control room.

The implementation has been validated with focused integration, recovery, chaos, and short real-provider staging tests. This documentation does not promise a universal throughput number: every deployment must validate its own provider, worker, database, Redis, connector, and concurrency limits before broad rollout.

How To Use This Documentation

This documentation set is written to replace guesswork with product truth. It covers what the app has now, how the main systems connect, what changed in the recent DeepSpace hardening phases, and where future development should stay careful.

If you are a user, start with Getting Started, Grounded Queries, Documents Hub, and Memory & Workspace. If you are developing the system, also read Connectors & MCP, Providers, Privacy & Security, and Architecture.

Durable-first DeepSpace

New chats no longer ask users to choose a runtime. Durable execution, restart recovery, replay, rehydration, and final-answer persistence happen automatically.

Phase 1

Long-session stability and safer runtime contracts are now implemented, including auto-compaction, compaction persistence, and reducer-safe mission state handling.

Phase 2

Deep observability and operator confidence are now implemented, including runtime diagnostics, hook and policy summaries, richer lane metadata, and canvas diagnostics.

Phase 3

This in-app documentation work is the practical Phase 3 layer for developer clarity, maintenance ease, onboarding, and safer future development.

MCP Strategy

The docs now explain what is already MCP-ready, what is still transitional, and how connector standardization fits the current architecture.

Editor + Workspace

The docs treat the note editor and workspace as core product surfaces, not side utilities.

Provider Clarity

Cloud, local, OpenRouter, embedding, reranker, and web-search runtime responsibilities are documented in one place.

What The Documentation Covers

The documentation now covers:

  • document ingestion and grounded query behavior
  • DeepSpace chat, streaming answers, and safe approval prompts
  • note editor, research tasks, exports, and working surfaces
  • persistent memory and conversation history
  • connectors, MCP transition direction, and external system handling
  • provider routing across cloud and local runtimes
  • privacy, secret handling, tenant boundaries, and admin posture