Use TAI from inside ChatGPT, Claude or Gemini
Your team already talks to an AI assistant. TAI has an MCP server, the open standard those agents use to connect to systems, and through it your assistant queries employees, time clock, payroll, tickets and indicators in plain language, with the permissions of whoever connected it. No spreadsheet exports, no passwords handed over.
The agent sees exactly what you see. Not one screen more.
Why your team's assistant knows nothing about the company
The data is locked in the system. To ask how many people are on vacation, someone opens the screen, exports and pastes into the chat. With yesterday's data.
Pasting a spreadsheet is a leak. Personal IDs, salaries and health data leave the platform for a chat that knows nobody's permissions.
Every integration is a project. API, key, code, someone to maintain it. The assistant waits for IT.
Nobody knows what the agent did. If it changed something, where was it recorded? With which permission?
One MCP server: Tali's own catalogue, permissions and executor
MCP is the open standard AI agents use to connect to systems, and TAI has its own. It has no logic of its own: it reuses Tali's operation catalogue, permission filter and execution. What your agent sees is what you would see on screen.
MCP for the agents on the market
ChatGPT, Claude, Gemini and any agent that speaks MCP connect to the same TAI server. You do not wait for a custom integration for each assistant.
Authorisation on a TAI screen
In assistants with connectors, such as ChatGPT and Claude, the connection opens a TAI screen asking whether you authorise it. You log in and click Authorise. No token, no shared password.
Token for command-line and in-house agents
For Claude Code, Gemini CLI or your company's own agent, create a personal token under Profile → Access tokens, with an expiry and with or without write access. Up to five active tokens, revocable instantly.
Read by default, write if you allow it
Without Allow write, the connection is read-only. With write, the agent creates and changes records within your permissions, with approval in your own assistant. Irreversible acts are always excluded.
The questions that today become exports
Everything Tali answers on the platform, your assistant answers from outside. The only difference is where the conversation happens.
Generated files, such as a payslip, a time sheet or a spreadsheet, arrive through a temporary link that expires in a few minutes. Expired? Ask again.
Examples
People and structure
"Who is on vacation this week?", "How many people are in sales?", "Who is Ana's manager?"
Time clock and payroll
"Who on my team has a pending clock correction?", "What is missing to close payroll?", "Generate my August payslip."
Service desk and processes
"Which tickets breach the SLA today?", "Where is my reimbursement request?", "Open a ticket for IT."
Finance and indicators
"What do I have to pay this week?", "How are the quarter's objectives?", "Who is off target?"
Connected apps and tokens appear on the same profile tab, with last use and expiry. Disconnecting takes effect immediately.
Where to connect
One MCP server, several agents, one rule: the permission is the connecting person's.
ChatGPT
Connector with authorisation on a TAI screen. Sensitive data such as personal IDs, health and bank details stay out of that channel by default.
Claude
claude.ai and Claude Desktop through a connector with authorisation; Claude Code through a personal token.
Gemini
Connection through the same MCP server, with a personal token with a set expiry, read-only or with write access.
Your own agents
Your company's agent, on any platform that speaks MCP, uses the server URL and a token.
Tali, inside the platform
To actually execute, including what is irreversible, Tali is the agent. External ones read and, if allowed, write.
Who it is for
- Executives who live in the AI assistant: Ask about the operation from where they already are, without opening one more system, and see only what the role allows.
- Companies with their own agent: The data or IT team connects the in-house agent to TAI by token and builds whatever they want on top, with no custom integration.
- Managers who want the answer now: Team clock, queue tickets, quarter objectives. The question leaves the phone and the answer comes back with today's data.
Questions about the connection
Which AI agents does TAI integrate with?
Does the external agent see everything in the company?
Can it change data?
Why can't ChatGPT see personal IDs and salaries?
I lost the token. Now what?
Does this replace Tali?
Connect your assistant and ask the first question
In 30 minutes, we connect your team's ChatGPT, Claude or Gemini to a demo account and you ask whatever you want about the operation.