v6.9.0 - Universal OpenAI-Compatible Speculative Anti-Refusal Gateway

Release v6.9.0 Highlights

• Universal OpenAI-Compatible API Gateway (START_GATEWAY.cmd / gateway.py): Local zero-dependency REST/SSE proxy server on http://127.0.0.1:8088/v1 that seamlessly works with ANY software (Cursor, ChatGPT Desktop, LibreChat, NextChat, Cherry Studio).
• Live Streaming 3-Token Speculative Interception: Automatically aborts refusal streams in < 0.03s and streams Model B's executive C++ deliverable directly to client UI.
• 100% Tested across 37 Unit Tests & 9 Release Steps.

v6.8.7 - Sane Relay-Friendly Reasoning Effort & Anti-Timeout Hardening

Release v6.8.7 Highlights

• Relay-Safe Reasoning Default (medium): Optimized default reasoning effort from high to medium, ensuring instant stream response (<2s) that completely avoids strict proxy timeouts.
• Automatic Auto-Repair Adaptation: Synchronized across repair.ps1, config.portable.toml, and turbo.ps1.

v6.8.6 - Relay Anti-Timeout Optimization & Anti-Buffering Headers

Release v6.8.6 Highlights

• Relay Proxy Anti-Timeout Optimization: Added X-Accel-Buffering: no, Cache-Control: no-cache, and Accept: text/event-stream headers to prevent Nginx/Cloudflare relays from buffering SSE chunks.
• Extended Stream Timeouts: Increased stream probe and executive generation timeouts up to 180 seconds to accommodate slow upstream reasoning models.
• 100% Tested across 9 Release Steps.

v6.8.5 - Fully Automated Silent Self-Healing on Every Launch

Release v6.8.5 Highlights

• Zero-Friction Automated Self-Healing: Every launch of TURBO.cmd, ADAPT_ALL.cmd, or �dapt.ps1 now silently self-heals corrupted TOML configs, restores desktop sessions, and validates environment dependencies upfront.
• 100% Sane Defaults & Zero User Action Required: Users never need to manually diagnose or repair syntax errors—the system heals itself on the fly.
• Verified Across Full 9-Step Release Suite.

v6.8.4 - 1-Click REPAIR_ALL Self-Healing Engine & Desktop Auto-Repair

Release v6.8.4 Highlights

• 1-Click Self-Healing Wizard (REPAIR_ALL.cmd / repair.ps1): Double-click to automatically fix broken TOML structures, clear toxic provider locks, and restore ChatGPT/Codex Desktop session history.
• ChatGPT Desktop Auto-Sanitizer: Automatically removes incompatible custom model provider tags that cause 'Model provider custom not found'.
• Universal Git Immunity: Guaranteed zero-crash execution across both Git and non-Git workspace folders.

v6.8.3 - Critical Fix for Non-Git Workspaces & Clean TOML Parsing

Release v6.8.3 Patch Highlights

• Non-Git Directory Fix (install.ps1): Prevented unhandled git rev-parse exception when target project is not a git repository.
• Clean TOML Root Key Sanitization ( urbo.ps1): Guaranteed �pproval_policy = never is always positioned at root level, fixing invalid type: string never in tui.model_availability_nux.
• Multi-Agent Universal Support: Reaffirmed native support across Codex CLI, OpenCode, Claude Code, Cursor, and ChatGPT.

v6.8.2 - Pure-Python .env Auto-Loading & 1-Click Relay Config Wizard

Release v6.8.2 Patch Highlights

• Zero-Dependency .env Auto-Loading: Pure Python parser automatically reads MODEL_A / MODEL_B endpoints and API keys with zero manual flags.
• 1-Click Relay Wizard (CONFIG_RELAYS.cmd): Double-click wizard to set independent A/B relays in 10 seconds without third-party proxy aggregators.
• Fallback Defaults: Automatically falls back to single global OPENAI_API_KEY when separate relays are omitted.

v6.8.1 - Maximum Fault Tolerance & Anti-Crash Launcher Hardening

Release v6.8.1 Patch Highlights

• Zero-Flashing Guarantee (TURBO.cmd / ADAPT_ALL.cmd): Added absolute PowerShell fallback paths and persistent pause on any error condition.
• Unextracted ZIP Guard: Automatic detection and actionable user prompts when double-clicked inside Windows compressed folder view.
• Semantic Versioning Policy: Stabilized release rhythm following SemVer patch cadence (v6.8.x).

v6.8.0 - Native macOS & Linux Shell Installers and Cross-Platform MCP

Release v6.8.0 Highlights

• Native macOS & Linux Scripts (install.sh, ADAPT_ALL.sh, TURBO.sh): Direct 1-line bash setup on macOS Terminal, Linux, and WSL.
• Standardized Model Context Protocol (MCP) Server: Pure-Python JSON-RPC 2.0 MCP server natively recognized by macOS Claude Desktop, Cursor, OpenCode, and Windsurf.
• Zero External Dependencies: Pure Python runtime ensures 100% out-of-the-box execution on macOS (Apple Silicon/Intel) and Linux distributions.

v6.7.0 - Standardized Model Context Protocol (MCP) Server & Multi-Agent Bridge

Release v6.7.0 Highlights

• Official Model Context Protocol (MCP) Stdio Server (mcp_server.py): Built-in JSON-RPC 2.0 MCP server natively exposing all
  e-toolkit tools to OpenCode, Cursor, Claude Desktop, Windsurf, and VS Code.
• Zero-Config Agent Discovery (.mcp.json): OpenCode, Cursor, and Claude Desktop automatically detect and register
  e-toolkit capabilities upon opening the workspace.
• Universal CLI Presets & Exporters: Instant output formatting for OpenAI ChatGPT Custom Instructions, OpenCode, and IDE agent rules.