How to share docs, tasks and memory with AI agents
Four steps: one place for the sources, one Doc, one task, one memory.
Summary
To share docs, tasks and memory with AI agents, keep them in one shared workspace instead of a chat thread. Put the source files and notes in the project the agent works in, let the agent write its result into a Doc you can both edit, track each request as a task that waits for your review, and let the agent remember durable facts so the next request starts with them. In Denker, that workspace is a project: its wall holds the files and Docs, the Taskboard holds the tasks, and memory holds what Denker learned.
Key points
- Docs: drop files, images and notes onto the project’s wall, and have the agent write its draft into a Doc that you edit directly.
- Tasks: requests that run as tasks show on the Taskboard with their runs, and a finished run waits for you to accept it or mark it not done.
- Memory: Denker learns facts from your chats or when you say “remember…”, and you can search, edit or forget them.
Most people share context with an AI agent by pasting it into a chat. That works once. The next day the docs are buried in old messages, nobody knows which request finished, and the agent has forgotten the preferences you explained. Sharing docs, tasks and memory means giving them a place that outlasts the conversation.
The short answer
Use a workspace that you and the agent both read and write. It needs three things: shared material (the docs and files the work is based on and the result being written), shared task state (what was asked, what ran and what is waiting for a decision) and shared memory (facts that should carry into the next request).
Denker is built around that idea. Each project has a wall of cards for files, images, notes and Docs, a Taskboard for tasks, and memory that Denker keeps across requests. The rest of this guide shows how to set each one up.
1. Share docs: put the sources where the agent works
Create a project for the work and drop the files, images and notes onto its wall. Then point your request at them: name the audience, the outcome and the material to use. Denker works from what your request points it to; what it can open beyond that depends on the tools and app access set up for the task.
Ask for the result as a Doc. The draft lands on the same wall as its sources, and it is an ordinary editor for you too: type, add headings, tables or charts from the slash menu, or select a passage and ask Denker to explain or rewrite it in place. There is no copy-paste step between the agent’s answer and your document.
2. Share tasks: make every request visible on a board
Requests that run as tasks show on the Taskboard as a board, list or calendar. Each task keeps its description, status, priority, due date and blockers, and its details list every run: when it started, how long it took and how it ended.
When a run finishes, the task waits for you. Accept the result, or mark it not done and say what is missing. Leave open questions as comments so whoever picks the task up next sees them first. Recurring work, like a weekly update, can run on a schedule.
3. Share memory: let the agent keep what it learned
Some facts matter beyond one task: who your customers are, how long updates should be, which languages they go out in. Denker learns facts like these from your chats, or when you say “remember…”, and recalls them in later requests.
Shared memory only helps if you can check it. Learned facts show up as cards on the wall, where you can hide or forget them, and the Memory frame lets you search and edit what Denker knows, accept or reject proposed changes and see where each fact came from. The knowledge graph is part of Pro.
4. Come back to the same project
The test of a shared setup is the next day. Open the project and the sources, the reviewed Doc, the task with its open comment and the remembered facts are where you left them. Ask for the next change in the same Doc instead of rebuilding the context in a new chat.
A one-off question still fits a chat. Work that spans several sources, drafts and decisions is what benefits from shared docs, tasks and memory.
What you need to set it up
Denker runs its agents through your configured Claude Code, Codex or OpenCode setup, and agent runs happen in the desktop app on Mac or Windows. You can open the same workspace in a browser while the desktop app is running.
Frequently asked questions
How do I share docs with an AI agent?
Put the files and notes in the workspace the agent works in rather than pasting them into a chat, and have the agent write its result into a document you can edit. In Denker, you drop them onto a project’s wall and the draft arrives as a Doc beside them.
How do I share tasks with an AI agent?
Run requests as tasks on a shared board. In Denker, each task shows on the Taskboard with its runs, and a finished run waits for you to accept it or mark it not done.
How do I give an AI agent memory across sessions?
Use an agent that keeps facts between requests and lets you review them. Denker learns facts from your chats or when you say “remember…”, recalls them later, and lets you search, edit or forget them.
Is a shared workspace better than a chat for this?
For a one-off question, a chat is enough. For work that spans several sources, drafts and decisions, a workspace keeps the material, the task state and the memory where both of you can find them again.