Sonnet 4.6 and chat compaction for compliance
ISMS Copilot uses Claude Sonnet 4.6 in Fast mode and Claude Opus 4.6 in Think mode. These models power the platform's compliance reasoning, delivering accurate control mappings, policy generation, and gap analysis for ISO 27001, SOC2, GDPR, and other frameworks.
Sonnet 4.6 improvements for compliance
Claude Sonnet 4.6 brings significant enhancements for compliance work:
1M token context window — analyze up to ~1,500 pages or multiple large documents in a single conversation
Superior reasoning — nuanced analysis of complex requirements like ISO 27001 Annex A controls and SOC2 trust principles
Zero control fabrication — tested extensively to eliminate hallucinations on framework controls and audit requirements
Audit-ready outputs — structured responses suitable for compliance documentation and auditor review
These improvements make Sonnet 4.6 reliable for iterative compliance tasks like multi-control mapping, policy refinement across versions, and detailed gap analysis.
Fast mode uses Claude Sonnet 4 for speed. Think mode uses Claude Opus 4.5/4.6 for deeper reasoning. Learn more in Choosing the Right AI Model.
How chat compaction works
Chat compaction keeps long compliance conversations accurate by intelligently summarizing older messages when threads approach ~150,000 tokens (roughly 75-100 exchanges). This feature is available in Think mode only.
The encrypted compaction workflow:
When your conversation nears the token limit, ISMS Copilot automatically triggers compaction
An amber banner appears: "Compacting our conversation so we can keep chatting…"
The system summarizes older messages (3-8 seconds), preserving key compliance details like control references, gaps identified, and policy decisions
Recent exchanges remain unchanged with full context intact
The conversation resumes, enabling indefinite threads for iterative compliance work
All compaction happens server-side with end-to-end encryption. Your data stays in EU-hosted infrastructure (Frankfurt) and is never used for AI training.
Compaction maintains accuracy by keeping your most recent detailed exchanges while summarizing earlier context. If you notice any loss of specificity, reference the control or requirement again in your next message.
Resetting context for new topics
To start fresh when switching compliance topics or frameworks:
Start a new conversation — recommended for new projects, frameworks, or unrelated compliance questions. This gives you a clean context window.
Use "Reset thread" — if a conversation stalls (no updates for 2+ minutes), click the "Reset thread" button to clear timers and fetch fresh messages. This doesn't delete history, only resets the connection.
For Fast mode (no compaction), new conversations are required when you reach the 300,000 token limit on the free tier.
For details on managing token usage and conversation limits by plan, see Managing Long Conversations and Usage.