AI that never invents a price, only the number your catalog gives it.

Managers stay in the CRM while AI drafts replies grounded in the live conversation and catalog, and — where the owner enables it — an in-CRM AI assistant answers questions and prepares actions across the workspace.

• AI-draft replies in the unified inbox, grounded in conversation history and real prices and stock
• Optional in-CRM AI assistant that reads your data and proposes changes for a human to confirm
• Every AI action runs through the same permission-checked operations, with an audit trail

Obriym exposes an official Model Context Protocol server, so you run the workspace from Claude, ChatGPT, Gemini, or Cursor without opening the UI — “show overdue deals and create a follow-up task for each” just works.

• Official MCP server over the same scoped Bearer tokens as the Public API
• OAuth 2.1 one-click connect; your workspace role is projected onto the tools you get
• Read tools run freely; every write is a confirm-before-run proposal
• Requires a Team or Scale plan, like the Public API it is built on

The first message in a messenger or marketplace chat can be handled by AI that answers from your catalog (price, availability, delivery), qualifies the lead, and hands off to a manager — approve-mode by default, autonomous only when you turn it on.

• Approve-mode by default: a human reviews and sends every reply
• Optional autonomous mode with a shadow-first rollout, hard caps, and low-confidence handoff
• EU AI Act disclosure on the first autonomous reply

AI Reply Copilot

An “AI draft” button in the reply box generates a suggested reply grounded in the real conversation history and the matching catalog products. The model replies in the customer's language and never invents prices or products. A human reviews and sends every draft.

• Grounded in conversation history plus live catalog (actual prices and stock)
• Human reviews and sends every draft
• Paid-plan capability, metered on monthly AI credits

First-line AI Seller

For approved channels, AI can prepare the first reply to an inbound customer message — price, availability, delivery — and qualify the lead. Approve-mode ships a draft for a human to send; an autonomous mode is planned.

• Approve-mode by default; autonomous mode is opt-in and shadow-tested first
• Hard per-conversation and per-day caps; low-confidence questions hand off to a human
• Answers only from the catalog; a request for a person always forces a handoff

AI copy for social media

In Publications the AI writes the post's base text, hashtags and a separate variant for each enabled network. It works from your catalog's real product data — name, price, availability — and is not allowed to invent a price or a discount. The result lands in the editor: you edit and publish it yourself.

• Grounded in the product card: no invented prices, stock levels or delivery promises
• A separate variant per network, truncated to that network's real character limit
• One generation = one AI credit; a human always publishes

In-CRM AI Manager

A session-scoped chat assistant that answers questions and prepares actions across the workspace over the same operations the MCP server exposes. It is shipped but off by default: it is enabled per deployment, and it is additionally gated by plan and AI credits — it is not enabled for every workspace automatically.

• Reads leads, deals, orders, products, reports; renders tables and clickable record links
• Every change is a confirm-before-run proposal, never applied silently
• Persisted chats, quick-reply options, and an audit trail per turn

MCP Server

The official Obriym MCP server lets AI clients (Claude, ChatGPT, Gemini, Cursor) operate your workspace through role-projected tools. It is the same shared operations layer as the Public API, behind OAuth 2.1 or a scoped Bearer token, and is listed in the public MCP registry.

• Read and guarded-write tools projected from your workspace role
• OAuth 2.1 discovery and one-click connect, or an existing scoped API token
• Per-tool scope checks and rate limits; no direct database access
• Paid-plan capability — requires Team or Scale; a Free workspace has no API or MCP access

Marketing-site AI agent

The assistant on this public site answers product questions from a fixed set of marketing facts and can capture a feature request as a lead. It has no access to any customer's CRM data and is rate-limited against abuse.

• Grounded only in public marketing facts, with no CRM read tools
• Can capture a feature request into the Obriym workspace
• Per-IP and global rate limits; loads only after cookie consent

Stock-grounded answers

Every customer-facing AI answer about price or availability is resolved deterministically against your real catalog and live stock before the model writes a word — so the AI states verified facts instead of guessing, and asks a clarifying question when it cannot resolve them.

• SKU and product-name resolution against the live catalog
• Verbatim price and availability facts handed to the model
• Clarify-first when a product or price cannot be resolved