Appearance
Understanding Context Usage
Quick Answer
The context meter shows how much of the AI's "memory" you're using in the current conversation. Hover over the meter to see exact token counts and access two in-thread controls: Compact context (summarizes history to free space) and Clear context (empties the thread entirely).
What Is Context?
Context is everything the AI assistant can "remember" during your conversation:
- Your previous messages
- Assistant responses
- Search results
- Analytics data
- Attached files
The Context Meter
What It Shows
The meter is a small circular progress ring near the chat input area. Hover over it (or tap on mobile) to open a panel that shows:
- Current token count and context limit (e.g.,
65.2K / 128K) - Percentage used
- Context management actions (Compact and Clear)
Reading the Meter
| Level | Range | Meaning |
|---|---|---|
| Normal | 0-70% | Plenty of room, assistant has full context |
| Warning | 70-85% | Getting fuller; consider compacting |
| Critical | 85-100% | Near limit; compact or clear recommended |
The ring color shifts from cyan (normal) to amber (warning) to red (critical).
Token Display
Token counts shown in the panel are real counts reported by the model after each turn, not estimates. There is no client-side approximation — the meter shows only the provider's exact usage for the latest turn. The count seeds correctly when you switch threads or refresh the page.
Because the number is provider-reported, the meter appears only once the model has reported usage for the current thread. On a brand-new thread — and right after you Clear context — the ring is hidden until the next assistant turn reports a token count.
How Context Fills Up
What Uses Context
- Every message you send
- Every assistant response
- Search results (especially large ones)
- Analytics data
- File contents
Approximate Usage
| Content Type | Context Usage |
|---|---|
| Short message | Small |
| Detailed response | Medium |
| Search with 20 locations | Large |
| Full analytics report | Large |
| Uploaded document | Variable |
What Happens When Context Is Full
Automatic Management
When a turn would overflow the context window, the agent automatically summarizes the older portion of the conversation into a compact digest before sending. The most recent 10 messages are always kept verbatim. This summarization is transparent — you can continue the conversation and the assistant retains the gist of what came before.
If auto-summarization is unavailable for a turn, the system falls back to trimming the oldest messages.
What Is Preserved
- The 10 most recent messages (kept verbatim)
- The assistant's knowledge of the summarized portion (condensed)
- Current search results in cache
What May Be Lost
- Verbatim detail in older messages (replaced by a summary)
- Very specific phrasing from early in the conversation
Managing Context Manually
The context meter panel (hover the ring) exposes two actions that let you control context proactively rather than waiting for the automatic threshold.
Compact Context
What it does: Summarizes the entire conversation history into a single compact digest. The thread keeps its knowledge with far fewer tokens.
How:
- Hover the context meter ring
- Click Compact context
- Confirm the dialog: "Compact conversation"
After compacting, you'll see a "Context compacted" confirmation. If the thread has no content old enough to compact, you'll see "Nothing to compact yet."
Important: The original messages are replaced and cannot be restored.
When to use: When the meter is amber or red and you want to continue the conversation without losing the thread's knowledge. Manual compact summarizes the full history (including recent tool results), which typically frees significantly more space than auto-compaction.
Clear Context
What it does: Empties the chat history for this thread entirely and starts fresh with no prior context.
How:
- Hover the context meter ring
- Click Clear context
- Confirm the dialog: "Clear conversation"
After clearing, you'll see a "Context cleared" confirmation.
Important: This cannot be undone. Your saved views and analyses are not affected — only the chat history is removed.
When to use: When you want a completely clean slate on the same thread, or when the conversation has gone in an unproductive direction and you want to restart without creating a new thread.
Availability
Both Compact and Clear are available whenever:
- The thread has at least one message
- No run is currently in progress
Both actions require confirmation before executing because they permanently alter the thread's history.
Managing Context Proactively
Keep Conversations Focused
- Stay on topic within a thread
- Start new threads for new topics
- Don't ask the same question repeatedly
Be Efficient
- Ask clear, specific questions
- Avoid very broad searches that return many results
- Summarize key findings if switching topics
When to Compact vs Start Fresh
| Scenario | Action |
|---|---|
| Thread getting long but still relevant | Compact context |
| Want to continue but meter is near critical | Compact context |
| Topic has changed completely | Start a new thread |
| Conversation went off track | Clear context or new thread |
Context Limits by Model
| Model | Context Size | Notes |
|---|---|---|
| Zeus.ai Secure (Third-Party) | 1,000,000 tokens | Default model |
| DeepSeek V4 Pro | 1,000,000 tokens | Max-thinking reasoning |
| Zeus.ai Clean Room | 1,000,000 tokens | 🔒 Clean Room subscription, text-only |
| Zeus.ai Clean Room (GPT-5) | 272,000 tokens | 🔒 Clean Room subscription, images/PDFs |
| Claude Opus 4.8 / Sonnet 4.6 | 1,000,000 tokens | Anthropic; images + PDFs |
| Claude Haiku 4.5 | 200,000 tokens | Anthropic; fast/cheap |
| GPT-5.5 / GPT-5.4 / GPT-5.4 Mini | 1,000,000 tokens | OpenAI; images + PDFs + URL fetch |
| Gemini 3.1 Pro / Gemini 3.5 Flash | 1,000,000 tokens | Google; full multimodal (audio + video) |
Tokens are roughly equivalent to words plus formatting. Context includes your messages, assistant responses, search results, and attached files.
Troubleshooting
Assistant Forgot Something
- The thread may have been compacted or trimmed
- Re-state the key information
- Or reference by specific ID or name
Context Meter Always High
- Long conversations fill up fast
- Use Compact context to shrink the thread while keeping knowledge
- Or start a new thread
"Nothing to compact yet" Toast
The thread's conversation history is too short or too recent for the summarizer to act on. Wait until more turns accumulate, or use Clear context if you want to reset entirely.
Compact Takes a While
Manual compaction sends the full conversation to a summarization model, which scales with thread length. Expect a few seconds to about a minute for a long thread. A spinner appears in the meter ring while compaction is in progress.