Conversations and projects

The agent workbench uses conversations to hold the context of one continuous task and projects to organize multiple conversations sharing a research goal. Splitting conversations and projects sensibly reduces subject mix-ups and makes questions, results, and files easier to find.

Know the conversation and project entries

The entries sit in the workbench left navigation.

Conversations and projects navigation

Area	Purpose
Start a new conversation	Create a new task that does not carry over the current conversation
Search conversations	Find existing conversations by keyword
Projects	Expand or collapse the project list
New project	Create a new research project
Recent	Show recently used conversations

The sidebar button at the top-left collapses the navigation to give the results area more room; select it again to restore.

A newly created conversation may first show the opening words of your question as a temporary title, replaced automatically once the system generates a formal one. For important tasks, rename afterwards to "subject + task type + key condition" for easier retrieval.

When to stay in the current conversation

Continue asking in the same conversation when you:

Add screening conditions to the current result.
Ask for an explanation of a parameter, score, or data source.
Change ordering, grouping, or output format.
Continue analyzing properties or designing validation around current candidates.
Correct the agent's understanding of the previous question.
Resume a task that was interrupted, paused, or failed—continue in the same conversation and pick up from saved progress without a new conversation or re-uploading attachments.

A running conversation keeps running when you switch away; on return, history is backfilled first and new content continues. Streaming content does not pile up in the background while you are away, avoiding a jumble on reopen.

When the concurrency quota is full, the page just shows the current queue or capacity state instead of repeating the same popup; queued states clear automatically after stopping. Do not resend the same question because of a capacity notice.

For example, after completing the PD-L1 × BMS-202 docking, you can continue with:

Compare only the poses that Vina and DiffDock returned successfully, listing input files, parameters, output files, and result limitations.

Follow-ups reuse the current context. When referring to "this protein", "the previous candidates", or "that structure", make sure the reference is unambiguous.

When to start a new conversation

Choose Start a new conversation when:

You begin studying an unrelated protein or target.
The new task should not inherit the previous task's assumptions and attachments.
You want to restart with a completely different analysis approach.
The conversation is too long to keep references clear.
The task belongs in another project or permission scope.

When starting fresh, pick an agent mode suited to the new task. Modes differ in workspace scope—confirm the mode covers the objects or data the task needs before starting. For complex research tasks, the system first asks you to confirm a plan and analysis outline, then executes.

In the new conversation, restate the needed protein names, database IDs, files, and constraints. A new conversation does not automatically receive all input files and specific requirements from the previous one.

Find past conversations

Select Search conversations to open a search window listing conversations by date.

Searching past conversations

Type keywords into the Search chats box.
Prefer distinctive content: protein names, database IDs, ligand names, or task types.
Wait for results to load, then select the conversation.
Close with Esc or by clicking outside the window.

For example, to find a docking task, search BMS-202, PD-L1, or 5IUS rather than just "analysis" or "protein".

Make conversations easier to search

After finishing an important task, name the conversation after its subject and goal—"PD-L1 + BMS-202 dual-tool docking" is easier to spot than "analysis results".

Manage a past conversation

Hover over a conversation under Recent and select the more-actions button on the right to open its menu.

The past conversation action menu

Action	Purpose	Suggestion
Rename	Change the display name in the list	Use "subject + task" naming
Move to project	File the conversation into an existing project or the unclassified area	Confirm the target project and memory scope first
Pin	Keep important conversations prominent	Good for tasks used often at this stage
Share	Open the conversation sharing flow	Check questions, attachments, and results for sensitive information first
Archive	Tidy unused conversations out of the active list	Confirm how retrieval works on the current platform first
Delete	Delete the conversation	Download result files you need to keep before acting

Rename a conversation

After selecting Rename, the name becomes editable.

Renaming a past conversation

Type the new name and select the check mark to save.
Select the X to cancel.
Renaming only changes the list label; it never re-runs the task.

Recommended naming:

subject + task type + key condition

For example:

PD-L1 + BMS-202 dual-tool molecular docking
CXCL12 human high-affinity receptor retrieval
PET hydrolase candidate screening

Before deleting

Deletion is a high-risk action. Check the conversation files panel first and download input files, results, and reports worth keeping; do not assume deletion is reversible.

Create a project

Projects organize multiple conversations under one R&D goal—for example, target research, candidate screening, structural analysis, and experiment planning each live in their own conversation, all filed into one project.

Select New project and fill in the basics.

The new project form

Field	Requirements and suggestions
Project name	Required, up to 50 characters on the current page; state the subject and stage
Description	Optional, up to 100 characters; state goal, scope, and key constraints
Memory scope	Required; decides how memory is shared between the project and outside chats

Select Create project when done.

Choose the project memory scope

When creating a project, choose between Project only and Global.

Choosing the project memory scope

Project only

The page describes it as: the project can access only its own memory, and its memory is invisible to outside chats.

Suited when:

Research context must stay inside one project.
Other ordinary conversations should not reference this project's memory.
The project involves an independent client, target, or internal R&D scope.

Global

The page describes it as: the project can access the memory of outside chats, and vice versa.

Suited when:

The project reuses general background from outside conversations.
Project knowledge should participate in other conversations.
Cross-conversation use of this information is confirmed to meet data governance requirements.

Memory scope is not a file-permission substitute

Memory scope affects how context is used between the project and outside chats, but it is not an organization permission, sharing permission, or data-masking mechanism. For internal data, also review account, organization, and sharing settings.

Whether the memory scope can be changed after creation follows the settings the project page currently offers. Decide isolation needs before creating.

Recommended project organization

Do not create a project for every question. A project corresponds to an ongoing research goal; conversations correspond to relatively independent tasks within it.