Home › Features › AI & assistants

Area 05

AI & assistants

Opelli gives AI the same connected business context your team works with: projects, tasks, time, customers, finance and published content. Use Opelli's built-in agents for focused suggestions, or connect your own assistant through MCP.

Every action follows the user's permissions, and connected apps can be revoked at any time.

Context in Markdown Opelli agents Bring your own AI assistant

Ask for it

The reply came in. Where does the deal stand?

Every mail, call and note on a customer sits on one timeline, with the deal on top of it. Ask what happened and your assistant reads the whole thread; ask for the next move and the deal is updated as your work, in the stage the whole team sees.

Context in Markdown

An assistant is only as good as what it can read.

Opelli writes in Markdown from the start — wiki pages, meeting notes, briefs and reports. The material an assistant needs is the material your team already wrote, in a format a model reads well.

  • Export your wiki without losing its structure — and import it back the same way
  • Give your assistant the same material your team already works with
  • Nothing is locked in: your content stays portable
The Audit module's Report view: a time range rendered as a Markdown document, grouped by module and project, with copy and email actions.

Opelli agents

AI prepares. You decide.

Opelli's built-in agents read what is already in the system — a customer conversation, project material — and come back with a suggestion next to the thing it came from. Nothing changes until you accept it.

A forwarded customer email on the customer timeline with a 'Read this' button and, below it, the suggestions: summary, next step, due date, signal, the amount the message mentioned, and the two people it named — one with an Add button, one already a contact.
  • Read a conversationGet a summary, the next step with its date, the amounts mentioned and the people named — ready to file on the customer.
  • Suggestions, not changesNo record is edited and no deal is moved for you. A drafted page is saved as an unpublished version — the live site changes only when you approve it.
  • Under your controlSensitive credentials are removed before content is sent to a model, and administrators can disable individual agents or the entire AI layer.

Administrators see a record of every model call.

The Agents activity log: one card per model call with the model and token counts, one payload unfolded to show what actually left, a run blocked by the private-key rule, and one whose password was removed before sending.

Ask for it

Last night, as one document.

Everything notable is recorded in one activity log across modules. Ask for a range and you get a single Markdown document with only the rows you are allowed to see — a ready starting point for a stand-up or a status update.

MCP server

Or connect your own AI assistant and just ask.

Opelli speaks the Model Context Protocol. Add Opelli as a connector in Claude, ChatGPT or another MCP-capable assistant and work with projects, tasks, planning, time, customers and published content — more than 390 tools, in plain language.

You

Catch me up on the latest customer conversation and update the opportunity when I ask.

search_customersupdate_deal
You

Turn a meeting note into linked tasks without copying decisions by hand.

get_wiki_pagecreate_task
You

Create a draft launch page from project material; the published version changes only when you approve it.

create_sitepublish_site
You

And update the Globex opportunity while you're at it.

customers · refused
Assistant

This connection does not have access to CRM, so I cannot update that opportunity.

Every tool call follows your own permissions: feature permissions, access limited to specific projects, and the rules for guest accounts. How to connect one is right below.

Connect an assistant

Your company's Opelli address, plus /mcp.

Every company's Opelli is its own MCP server, next to its own data. There is no shared endpoint. The server uses the same API as the web app, so an assistant gets exactly the permissions of the person who connected it, and nothing beyond them.

  • Claude, ChatGPT and other connectors. Add https://<your-company>.app.opelli.dev/mcp as a custom connector and sign in the way you sign in to Opelli. OAuth 2.1 with PKCE; registration and discovery are automatic, so there is no key to paste.
  • Scripts and local agents. A personal API key (opk_…) from Settings → Personal → My API keys, limited to the modules it may reach, sent as a bearer token.
  • Limits an assistant cannot argue with. A key acts as its owner and never reaches further. Administrators can switch MCP off for the whole company or allow only some modules, and external accounts cannot create keys at all.
Claude Desktop, with an API key 
{
  "mcpServers": {
    "opelli": {
      "command": "npx",
      "args": ["-y", "mcp-remote",
        "https://<your-company>.app.opelli.dev/mcp",
        "--header", "Authorization:${OPELLI_KEY}"],
      "env": { "OPELLI_KEY": "Bearer YOUR_KEY" }
    }
  }
}

The key goes in env rather than in args, because Claude Desktop breaks the space in Bearer <key>.

See it on your own operations.

Opelli runs in the cloud on an opelli.dev address or your own domain. Leave your email and we will show you how it fits the way your team works.

Previous← Security & control