Telegram front door A privacy-focussed channel for instructions, check-ins and approvals.
How it works
Share what helps. Keep control.
Colleague is not just a chatbot or a prompt window. It can access the internet, take action, work with files, remember context and check in proactively. It has chat, a dedicated Google Workspace account, memory, safeguards and frontier AI models, all built on top of OpenClaw.
Work with it like a colleague.
Each Colleague has its own Google Workspace account. You can share documents, folders, notes, calendars and forwarded email threads with it like a real collaborator. It can also work without Workspace access when the task does not need shared files or calendars.
Chat, workspace, action, intelligence.
The pattern is deliberately familiar: talk to Colleague, share useful material when the task needs it and let the agent do bounded work around the sources you choose.
Colleague Your agent
One assistant with its own workspace, memory, safeguards and clear data boundaries.
Telegram front door 
Google Workspace account Share files, folders, documents, calendars and email context like you would with a collaborator.
Ask questions, draft text and reason through work even before any files or calendars are shared.
Internet access, actions, memory, safeguards and proactive work, not just answers.
Frontier AI models Uses OpenAI’s most powerful models to read, reason, draft and decide what to do next.
The agent sees the sources you share, not everything by default.
Templates, edited drafts, meeting briefs, response letters and next steps.
Chat is the front door.
Telegram gives Colleague a familiar place to receive quick questions, drafts, reminders and approvals. It keeps the assistant close to the working day instead of hidden in another app.
Live documents stay live.
Work can happen in the files themselves: shared drafts, notes, plans, calendars and folders that stay visible to you as the project moves.
Share what helps.
Colleague has its own Google Workspace account, so you can share documents, folders, calendars or email context without connecting your personal accounts.
It can act, not just answer.
Colleague can access the internet, work with files, remember context, prepare drafts and check back in. The point is not another text box; it is an assistant that helps move work forward.
Memory gives continuity.
Colleague can carry useful context between sessions: active projects, decisions already made, recurring collaborators, preferences, deadlines and the shape of the work overall.
Control stays with you.
You decide what the agent can see by sharing specific material with its account, the same way you would with an external collaborator.
What goes to the model?
When a task needs AI inference, Colleague sends the user message, relevant working context, selected excerpts or action results and the instructions needed for the task. The working pattern is selective: share the document, folder, calendar or email context the task needs.
What can it do?
Because Colleague is built on OpenClaw, it can access the internet, take action, maintain memory, work with files, carry out multi-step tasks and check back in when something needs attention. That is the difference between a model answering a prompt and an agent helping move a project forward.