QVeris
Connect your AI

RELEASE NOTES

QVerisLab: Memory, Models and Tool Checks

Use historical notes as a checklist for memory, model switching and tool calls; verify current availability in QVerisLab.

Open QVerisLabOpen the evaluation prompt

Published: March 16, 2026 · By the QVeris Team

On this page
Dark blue concept visualization of conversation memory, a model-selection control connected to two model engines, and a release verification checklist.
Review the release changes, then verify memory retention, model switching, and tool-call behavior with repeatable workflow tests before adopting the update.

What's New in v2026.3.8

NEW

🧠 Persistent Memory Upgrade

The original article describes memory and summarization upgrades. Verify context limits against the actual release.

  • Verify the originally stated 32K-to-96K context change
  • Smart summarization preserves entities and decisions
  • Verify opt-in cross-session memory
  • Check /memory view and edit behavior
QVerisLab — Terminal
# New memory commands: /memory show # View current memory /memory clear # Clear session memory /memory pin "fact" # Pin a fact permanently
NEW

🔀 Model Switching

Test switching models mid-conversation for context retention and task fit.

Claude 3.7 Sonnet

Best for: reasoning & code

GPT-4o

Best for: general tasks

Gemini 2.0 Pro

Best for: multimodal

DeepSeek R2

Best for: cost efficiency

QVerisLab — Terminal
# Switch model in chat: /model claude # Switch to Claude 3.7 Sonnet /model gpt4o # Switch to GPT-4o /model gemini # Switch to Gemini 2.0 Pro /model deepseek # Switch to DeepSeek R2

Verify full history and QVeris tool access after switching.

IMPROVED

🔀 Smarter Capability Routing

The original article describes intent parsing and provider-selection changes; validate them with representative tasks.

  • Original claim: 23% better intent detection; reproducible evidence not supplied
  • Provider selection now weights recent success rates
  • Parallel tool calls for multi-step queries
  • Automatic retry with fallback provider on failure
  • Original claim: 180ms lower average latency; reproducible evidence not supplied

Before v2026.3.8

71%
Intent accuracy (original illustration)

After v2026.3.8

94%
Intent accuracy (original illustration)
IMPROVED

📋 Response Formatting

Check Markdown, tables and code-block display against expected behavior.

  • Auto-detects when to use tables vs prose
  • Code blocks now include language detection
  • Check financial table formatting
  • Long responses include automatic section headers
  • Copy button on all code blocks and tables

60+ fixes listed in the original article

The following is the original checklist; verify each item against your installed version.

Memory & Context15 fixes

  • Fixed memory duplication on session reload
  • Resolved context overflow causing truncation
  • Fixed pinned facts being lost after 24h
  • Memory export now includes full conversation tree
  • Fixed /memory clear not persisting across tabs
  • + 10 more fixes

Tool Calls & Routing18 fixes

  • Fixed race condition in parallel tool calls
  • Resolved auth token expiry during long sessions
  • Fixed JSON parsing errors on malformed API responses
  • Tool call retry now respects user timeout settings
  • Fixed duplicate tool calls on network retry
  • + 13 more fixes

Model & Response14 fixes

  • Fixed model switching losing system prompt
  • Resolved streaming interruptions on slow connections
  • Fixed code block rendering in markdown responses
  • Model preference now saved per conversation
  • Fixed response cutting off at exactly 2048 tokens
  • + 9 more fixes

UI & Experience13 fixes

  • Fixed dark mode flicker on page load
  • Resolved mobile keyboard pushing chat input off screen
  • Fixed copy button not working in Firefox
  • Conversation list now loads incrementally
  • Fixed search not finding older conversations
  • + 8 more fixes

Original release history

v2026.3.8
Mar 8, 2026
Memory upgrade · Model switching · 60+ bug fixes
ARTICLE VERSION
v2026.2.15
Feb 15, 2026
Original record: capability network expansion
v2026.1.20
Jan 20, 2026
Original record: public launch, providers and MCP support
v2025.12.1
Dec 1, 2025
Original record: private beta and routing engine

How to check the update

🌐

Web App

Open the app and check its current version and features. This article does not confirm a live update.

Open QVerisLab →

💻

API

For API integration, verify model parameters and available features in current official documentation.

View the REST API Reference →

🔌

MCP Server

Follow current MCP documentation and test routing and context behavior in Cursor or Claude.

QVeris MCP — Terminal
npx @qverisai/mcp@latest

MCP Docs →

Frequently Asked Questions

What's new in QVerisLab v2026.3.8?
The original article lists three areas:persistent memoryand expanded context,mid-conversation model switchingswitching across models, andsmarter capability routingplus memory, tool, model and interface fixes. Metrics and completion status require independent verification.
How does QVerisLab's memory upgrade work?
The original article describes context expansion from 32K to96K tokens (verify by version)with smart summarization that preserves key entities and decisions. Memory can now persist across sessions (opt-in), and new/memorycommands let you view, edit, pin, and clear stored context at any time.
Can I switch AI models in QVerisLab?
Yes.Use/model claude, /model gpt4o, /model gemini, or/model deepseekto test model switching, history retention and tool permissions.
How do I update to the latest version?
Check theweb appfor its current version. ForAPI users, consult current documentation. ForMCP Server, runnpx @qverisai/mcp@latestas documented and validate the result.
How many bugs were fixed in v2026.3.8?
Over 60 issues were listedacross memory/context (15), calls/routing (18), models/responses (14) and UI/experience (13). This article retains the checklist; the individual fixes have not been independently verified.

Check the release in your workflow

Use these historical notes as a test checklist; verify memory, model switching and tool behavior in your installed version.

v2026.3.8 · Historical notes
Open QVerisLab