v3.1.1

OpenSuiteMCP 3.1.1

Added

• /docs/upgrades — tag-based self-host update runbook (pnpm skills:sync, migrate, rebuild)
• Licensing clarity — free for your org’s internal self-host; paid delivery/support only via Unstacked Apps (self-host docs, README, NOTICE)

Fixed

• pnpm dev defaults to webpack so Turbopack no longer fails on @ai-sdk/provider-utils dynamic imports. Optional: pnpm dev:turbo

Upgrade notes

1. Pull v3.1.1 (or main) and use pnpm dev as documented
2. After go-live, see Upgrades (also /docs/upgrades when self-hosting)

See CHANGELOG.md for full notes.

v3.1.0

OpenSuiteMCP 3.1.0

Added

• NetSuite connect wizard — guided App Portal flow, shared Integration checklist, manage-accounts polish
• Public docs — /docs, /docs/netsuite-integration, /docs/self-host (org-hosted architecture overview)
• Oracle skills sync — pack no longer vendored; run pnpm skills:sync (or rely on production entrypoint) to populate .data/oracle-skills
• Shared model registry — canonical Speed / Reasoning models for Google, Anthropic, OpenAI

Fixed

• Anthropic Sonnet thinking uses adaptive + effort (Sonnet 5+ rejects thinking.type.enabled)
• Mobile dialog/sheet footer gaps and chat-history tooltip collisions

Upgrade notes

1. After pull: pnpm install then pnpm skills:sync before relying on Oracle skills in the portal
2. Optional: schedule weekly pnpm skills:sync (or cron) for new upstream skills
3. Set AUTH_URL so /docs/netsuite-integration shows your real OAuth redirect URI

See CHANGELOG.md for full notes.

v3.0.0 - App Portal, Skills, Prompts & NetSuite MCP

App Portal

One place for Chats, Skills, Prompts, AI Provider, NetSuite, Web Search, Timezone, and Account — Claude-style left rail with a consistent panel layout.

SuiteCloud Agent Skills

• Oracle SuiteCloud Agent Skills pack plus custom SKILL.md skills
• Enable/disable per session; enabled skills inject into the system prompt for new messages

Companion Prompt Library

• Native browse of Companion SuiteApp templates (category / industry / role filters)
• Fill placeholders, then send the prompt into chat

NetSuite MCP

• Multi-account management with DCR probe and clearer Integration setup
• Connection status chip and improved connect / disconnect flow

BYOLLM & usage limits

• Providers: Google (Gemini), Anthropic (Claude), and OpenAI (GPT) with your own API keys
• Daily message caps via MAX_MESSAGES_PER_DAY_REGULAR / MAX_MESSAGES_PER_DAY_GUEST
• Optional Redis burst limit via CHAT_BURST_LIMIT_PER_MINUTE (fail-open if Redis is unset)

Breaking changes

• Inception Labs removed — migrate to Google, Anthropic, or OpenAI
• Web search narrowed — Folio3 and Tim Dietrich tools removed; NetSuite Help Center search remains
• Run new migrations: 0007_netsuite_accounts_dcr, 0008_user_skill_settings

Technical

• package.json version set to 3.0.0
• Skills and NetSuite account settings persisted in the database

v2.6.0 - Custom Instructions, Prompt Enhancements & Migration Fix

Custom Instructions (instructions.md)

Add your own instructions to shape how Ava responds.

• Settings → Custom Instructions — Import an instructions.md file or paste content
• Custom instructions are appended to the system prompt as "Additional User Instructions"
• Protected core directives — Tool completion, no fabrication, orchestration rules, and Ava identity cannot be overridden by user instructions

System prompt improvements

• Refined identity, response rules, search triage, and tool orchestration
• Intent-based search triage instead of fixed priority
• Step budget clarified: "Do not stop early unless the objective is satisfied"
• Search scaling: prefer 1–2 targeted searches; add more only when addressing clearly distinct sub-topics and staying within the step budget
• Explicit guidance for fiscal/quarter-based queries to derive period from the provided date

Tool orchestration updates

• Resolution model: Fully Resolved, Partially Resolved, Blocked
• MCP rules: max 3 consecutive calls before alternating with search
• Completion: stop only when objective is met, NetSuite operation completes, or the system ends the turn
• Protected directives prevent user instructions from overriding safety and orchestration rules

Bug fix: migrations (inceptionApiKey)

2.4.0 added inceptionApiKey to the schema and journal but the migration file was never committed. 2.5.0 did not fix this. Users on 2.4.0/2.5.0 could see "column already exists" or missing-column errors.

Migration 0006_illegal_sunfire now adds inceptionApiKey and customInstructions with IF NOT EXISTS for reliable fresh installs and upgrades.

Technical

• New customInstructions column in UserSettings
• PROTECTED_DIRECTIVES in prompts.ts enforce non-overridable core rules when custom instructions are present

v2.5.0 - Inception Labs Mercury 2 Enhancements

Inception Labs provider enhancements

This release improves the Inception Labs (Mercury 2) integration with a custom provider and streaming experience.

✨ What's new

• Custom Inception provider — Dedicated provider (lib/ai/custom-providers/inception.ts) with reasoning summary extraction for Mercury 2
• Diffusion streaming — Live refinement display during reasoning mode so you can see output as it’s generated
• Extended model options — Reasoning effort, summary, and diffusing parameters forwarded to the Mercury 2 API

🧰 Technical

• Added @ai-sdk/openai-compatible as an explicit dependency for Inception Labs (Mercury 2) integration

v2.4.0 – Inception Labs Mercury 2 Integration

✨ Added

• Inception Labs provider with Mercury 2 for both speed and enhanced reasoning.
• API key support for Inception Labs in Settings (encrypted at rest).
• OpenAI‑compatible chat completions integration for Mercury 2.
• Model selector updates to include Mercury 2 (Speed + Enhanced Reasoning).
• Tooling updates so MCP tool metadata generation uses the selected provider.

🔧 Updated

• Provider-aware configuration across chat, settings, and tool outputs.
• Documentation and app metadata to reflect the new provider.

✅ Notes

• Reasoning effort parameters are forwarded for Mercury 2 reasoning mode.
• Migrations included for new inceptionApiKey storage.

v2.3.0 – Custom Web Search Tools

Custom Web Search Tools

This release replaces the single web search + list-search-domains flow with three domain-specific search tools so Ava can target the right sources:

• NetSuite docs – Official Oracle NetSuite Help Center (foundational docs, UI, permissions, SuiteScript API)
• Tim Dietrich – NetSuite developer blog (SuiteQL, SuiteScript, financial prompts)
• Folio3 – NetSuite integration & MCP (MCP setup, API workflows, reconciliation)

Highlights

• Settings → Custom Web Search Tools – Per-domain toggles control which search tools are available in chat. Only enabled tools are offered to the model.
• System prompt triage – The model is guided on when to use each search tool.
• Optional Redis cache – 7-day TTL for repeated queries per domain (set REDIS_URL to enable).
• Migration safety – 0005_dark_the_order now uses IF NOT EXISTS for maxIterationsReached and maxIterations for safe re-runs.

Also in this release

• Tool status badges support "Approval requested", "Approval responded", and "Denied".
• Get current config shows OpenAI provider with emerald styling.
• Upgraded @ai-sdk/react to ^3.0.26 (ai@6.0.50 via pnpm overrides).

Full changelog: CHANGELOG.md

v2.2.0 – Iteration Management System and Workflow

What's new

Iteration Management System and Workflow

• Flag-based max reasoning steps – No in-thread message injection; a database flag controls when the thread is locked.
• Configurable limit – Max reasoning steps (1–20) in Settings → AI Provider, default 10.
• Info card when limit is reached – Card appears above the input with three options:
  • Check NetSuite KB and continue – Clears the flag and sends a message to search NetSuite web resources, then continues.
  • No, I'm fine – Clears the flag and unlocks the thread.
  • Brute force it – Clears the flag and sends a message asking the assistant to continue.
• Thread locking – Input stays disabled until the user chooses an option; the card persists on reload.
• API – GET /api/chat/[id] returns maxIterationsReached; POST /api/chat/[id]/max-iterations clears the flag.
• Database – Single migration adds UserSettings.maxIterations (default 10) and Chat.maxIterationsReached (default false).

---

Full changelog: v2.1.0...v2.2.0

v2.1.0 - OpenAI Support & Error Handling Improvements

🎉 What's New

OpenAI Provider Support

We're excited to add OpenAI as a third AI provider option! You can now choose between Google Gemini, Anthropic Claude, and OpenAI GPT models.

• GPT-5 Mini for speed mode - Fast responses and efficient tool calls
• O4 Mini for enhanced reasoning - Advanced reasoning capabilities for complex agent tasks, chained tool calls, and structured data analysis
• Full integration with existing settings and model selection UI

Note: For OpenAI enhanced reasoning features, you'll need to verify your organization at platform.openai.com/settings/organization/general

Improved Error Handling

Errors are now displayed directly in the chat interface with better visibility and persistence:

• Error messages appear as formatted cards in the chat
• Errors persist across page reloads
• Improved dark mode styling for better readability
• Better error recovery - you can send follow-up messages immediately after errors

🐛 Bug Fixes

Settings Modal

• Fixed intermittent form clearing issues that caused fields to appear empty
• Fixed provider dropdown display when switching between providers
• Fixed skeleton loading states to properly match input field dimensions
• Removed double scrollbar and improved overall layout spacing

Error Recovery

• Fixed status reset after errors to allow sending follow-up messages
• Fixed empty message prevention after errors occur
• Fixed stream interference issues when sending messages after errors

🎨 UI/UX Improvements

• Simplified settings save mechanism (removed "Save and Edit" dropdown)
• Updated model descriptions to accurately reflect the models being used
• Improved error card legibility in dark mode
• Better spacing between header, form content, and action buttons in settings modal

📦 Dependency Updates

• Upgraded @ai-sdk/anthropic from v2.0.57 to v3.0.23
• Upgraded @ai-sdk/google from v2.0.26 to v3.0.13
• Resolved compatibility warnings for reasoning features

📚 Documentation

• Simplified README to focus on purpose and quick start
• Added documentation table of contents
• Added app screenshot and icon to README
• Removed redundant content (moved to dedicated documentation files)

🔧 Technical Improvements

• Comprehensive error handling with chat-related error filtering
• Error flag management to prevent stream interference
• Proper status management for error recovery
• Improved type safety in provider configuration

---

Full Changelog: See CHANGELOG.md for complete details.

v2.0.0 - Complete rewrite with Next.js and Vercel AI SDK

🎉 v2.0.0 - Complete Rewrite

Complete architectural overhaul from LangChain/Express to Next.js with Vercel AI SDK.

⚠️ Breaking Changes

This is NOT backward compatible with v1.x

• Complete rewrite - new codebase
• New database schema required (PostgreSQL with Drizzle ORM)
• New authentication flow (NextAuth)
• New API structure (Next.js App Router)

Major Changes

Architecture

• ✅ Migrated from LangChain/Express to Next.js App Router
• ✅ Replaced LangChain with Vercel AI SDK
• ✅ Migrated from file-based sessions to PostgreSQL with Drizzle ORM
• ✅ Added NextAuth for authentication

AI Providers

• ✅ Multi-provider support (Google Gemini, Anthropic Claude)
• ✅ Reasoning/thinking modes for both providers
• ✅ Speed vs Enhanced Reasoning model selection

Features

• ✅ Complete UI redesign with shadcn/ui components
• ✅ Chat history and persistence
• ✅ User management and authentication
• ✅ Settings management with encrypted API keys
• ✅ Title and summary generation for chats

License

• ✅ New Sustainable Use License with Unstacked Apps, LLC ownership

Migration Notes

Users upgrading from v1.x will need to:

1. Set up new database (PostgreSQL)
2. Run migrations (pnpm db:migrate)
3. Re-authenticate with NetSuite
4. Re-configure AI provider settings

Full Changelog

See CHANGELOG.md for complete details.