Changelog

This release adds the ability to attach existing workspace knowledge to agents and includes backend and widget fixes for reliability and correctness.

New features

• Add from Workspace — In the agent knowledge panel, you can now attach existing workspace items (documents, webpages, raw text) to an agent without re-uploading. Open “Add from Workspace,” select one or more items, and click Attach. The agent gains access to that knowledge immediately.

Improvements

• Chat widget: More reliable behavior when expanding on desktop and tablet, and improved handling of streaming responses.
• Notifications: More consistent API summary and notification loading in the notifications dropdown.

This release adds version history for memories and introduces support for external MCPs so the thinking agent can act on your tools and data.

Memory version history

Memories now keep a version history. You can see who changed what and when, compare versions with a diff (Git-style), and restore a previous version from the workspace UI or via agent tools.

• Version list and diff viewer in the memory card (“View versions”).
• Agent tools: get_memory_versions, restore_memory_version.
• Each create/update (from UI, agent, or MCP) stores a new version.

MCP support in Qoris

The thinking agent can use external MCPs so it can act on your tools and data. You attach these per project; the agent gets a single, coherent tool set.

• Linear — Issue tracking and project management; create, update, and search issues (remote MCP).
• GitHub — Repos and issues; create issues and PRs, search code, manage repositories.
• Notion — Knowledge base; search, read, and update Notion pages and databases.
• Postgres / Supabase — Connect your own Postgres or Supabase instance; the platform handles connection strings safely (including special characters in passwords).
• Custom MCPs — Add any MCP that speaks SSE or HTTP: provide a URL (and optional headers) and use it alongside the built-in integrations.

Why it matters

• Memories are auditable and reversible.
• The agent can integrate with your own databases and tools via external MCPs.

We’ve released our own MCP server so any compatible client can read and write your Qoris workspace memories and search your project knowledge.

What it is

The Qoris MCP is an HTTP MCP server at mcp.qoris.ai. It exposes your workspace as tools: save and retrieve memories, update and delete them (with scoped permissions), and search the project knowledge base. Clients connect once and then call these tools like any other MCP.

Authentication

You can use it in two ways:

• API keys — Create a project-scoped API key in the Qoris UI, choose scopes (e.g. save/get/update/delete memories, search knowledge), and use it as a Bearer token. Ideal for scripts, custom agents, and Cursor.
• Interactive OAuth — Use the standard OAuth 2.0 Authorization Code flow with PKCE. Users sign in with Qoris, authorize the app, and get an access token. Supported by ChatGPT, Claude, and other clients that use OAuth for MCP.

Where you can add it

Add the Qoris MCP in:

• ChatGPT — As a custom app using the MCP endpoint and OAuth.
• Claude — Via Claude Desktop or other Claude clients that support MCP and OAuth.
• Cursor — Using the MCP config (URL + API key) from the Qoris workspace.
• Any MCP-capable client — Use the same HTTP JSON-RPC endpoint with an API key or OAuth token.

What you get

• Save memory — Persist facts and preferences for the workspace.
• Get memories — List and paginate all workspace memories.
• Update / delete memory — Change or remove memories (subject to ownership and scopes).
• Search knowledge — Query the project knowledge base from outside Qoris.

Summary

Memories created or updated via the MCP are visible in the Qoris UI and to the thinking agent; they support the same version history and diff/restore as the rest of the platform. The Qoris MCP turns your workspace into a reusable, cross-app memory and knowledge layer.

We’re excited to share a major milestone: a new set of capabilities across the platform, led by the Orchestrator, Memory, and a suite of supporting features designed to fundamentally elevate how your systems think, adapt, and perform.

This launch reflects a shift from tools that react to platforms that reason, remember, and orchestrate—so every engagement makes the system stronger.

The Orchestrator: intelligence in motion

A new core layer that brings structure, coordination, and decision-making clarity across your workflows. Rather than isolated tools operating independently, the Orchestrator acts as the central intelligence that coordinates tasks, agents, and processes; routes decisions based on context, priority, and intent; and ensures every component works together seamlessly.

• Dynamically coordinates tasks, agents, and processes.
• Routes decisions based on context, priority, and intent.
• Ensures that every component of the system works together seamlessly.

Memory: context that compounds over time

With QORIS Memory, your platform retains meaningful context, learns from prior actions, and continuously refines its understanding of your business—turning interactions into long-term intelligence.

• Persistent awareness across sessions and workflows.
• Smarter personalization and decision-making over time.
• Reduced repetition and friction for your teams and users.

Additional features for scale and trust

• New and more intelligent Widget UI.
• Marketplace of pre-built AI agents.
• Improved control and visibility across workflows.
• Greater flexibility in configuration and execution.
• Stronger foundations for scalability, security, and performance.

Why this matters

• Faster execution with less manual oversight.
• Systems that adapt as your needs evolve.
• A foundation built for long-term growth, not short-term fixes.