A saved conversation keeps its transcript, models, mode, prompts, generation settings, workspace folder, tool selections, skills, and Knowledge settings. Older messages remain visible even when they no longer fit in a model request. This page separates transcript history from the narrower context SoAI sends to a model.
In the SoAI WebUI, open Chat from the sidebar and select the conversation before using the header, composer, or Chat configuration paths below.
Context limits and compaction
| Context | Scope | Set it here |
|---|---|---|
| Message text or direct attachment | One message in one conversation | Composer and Add attachments |
| Conversation system prompt | Model requests in the conversation | Chat configuration > Parameters > System prompt |
| SoAI system prompt | Eligible SoAI-managed model requests | Chat configuration > Parameters > SoAI system prompt |
| Skills | Reusable instructions available to one ordinary chat | Chat configuration > Skills |
| Knowledge | Indexed documents for one conversation | Add attachments > Knowledge |
| Memory and profile | The signed-in account across chats | Chat configuration > Memory > Open profile form |
| Workspace folder | File base for one conversation | Chat configuration > Files > Change folder |
Lock system prompt carries the same system prompt into new messages and conversations until it is unlocked. Inherit default returns a conversation workspace to the account's default folder. A later default-folder change removes an override that falls outside the user's allowed workspace.
Watch context occupancy
Open Chat configuration > Appearance > Input actions, enable Token counter, and select Save. The composer counter shows measured or estimated prompt use, including the enabled SoAI system prompt, and the selected model's context window when SoAI knows it. Its color changes near the limit. Select it to switch between token occupancy and generation speed, or hold it to open Parameters with the context length in view.
Compact older context
Before an agent request, SoAI can compact older history when the full prompt no longer fits its available budget. To start it yourself, wait until the conversation has a message and is idle, then select Compact in the header. SoAI uses the selected conversation model, summarizes older context, and places a visible boundary in the timeline. Later requests use that summary instead of sending every earlier detail word for word. If the summary would not be smaller than the context it replaces, SoAI keeps the existing history and reports that nothing was compacted.
Expand a boundary to inspect its summary, injected prompt, and statistics. While idle, select Regenerate to rebuild a manually created boundary or Remove boundary to return the earlier transcript to normal context. Check occupancy before removing one because the restored detail can exceed the model's prompt budget.
Skill instructions use context too
Selected skills become instructions when the assistant loads them or you request them explicitly. Their instructions consume the model's context budget and remain available to the active work across compaction. If a skill cannot fit, it is not partially applied. Compact older history while idle, use a model with a larger context window, or shorten the skill. Long references belong in supporting files read as needed.
Accepted work retains the skill content selected when it was admitted, even if you remove the skill afterward. A regenerated reply uses the selection currently saved for the conversation. A copied transcript does not select skills in its new conversation; enable them there before continuing.
Messaging conversations compact independently
Automatic compaction also applies to messaging conversations. Each messaging account can set a different context length under the Settings button at the right of the header, then Settings > Messaging > Edit > Chat execution > Edit parameters > Model settings. Set Context length (tokens), or clear it to use the selected model's default. A smaller value makes that account compact sooner and does not change other messaging accounts.
Edit and manage conversation history
| Action | History effect |
|---|---|
| Edit | Replaces a user message, removes later messages, and generates a new continuation. Direct attachments on that message can be removed before saving. |
| Resend message | Resubmits that user message and removes all later messages before the new reply. |
| Regenerate | Replaces an assistant reply and removes messages after it. |
| Delete | Shows a temporary deleted placeholder with an undo period. Selecting Delete again commits immediately. |
History-changing actions are unavailable during a turn. Export or clone before rewriting a branch if later turns must be retained.
Clone, archive, and delete
Use a conversation row's Edit action to rename, recolor, favorite, archive, or export it. Use the sidebar selection toolbar for Archive selected, Clone selected, or Delete selected. Active or queued work must finish or be cancelled before cloning or deletion.
A clone receives its own transcript, settings, recorded tool activity, and ready direct-file attachments. Its indexed Knowledge collection is not copied, so Knowledge cards from the source transcript appear unavailable until you add or import those documents again. Skill selections are also not copied; select and save the skills the clone needs. Archive hides a conversation without deleting it; use Archived conversations to find and restore it.
Drafts, queues, and reconnection
SoAI saves composer text and ready direct attachments as a conversation draft and synchronizes them between signed-in browsers. If another browser changes the same draft while you are typing, SoAI keeps the local version and reports a conflict. Leaving Chat does not stop accepted work. On reconnection, the page reloads drafts, queued prompts, messages, tool activity, and attention state. Cancel a queued item before it starts; use Stop after it begins.